xnu-3248.20.55.tar.gz

[apple/xnu.git] / osfmk / kern / waitq.c

/*
 * Copyright (c) 2015 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */
/*
 * @OSF_FREE_COPYRIGHT@
 */
/*
 * Mach Operating System
 * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
 * All Rights Reserved.
 *
 * Permission to use, copy, modify and distribute this software and its
 * documentation is hereby granted, provided that both the copyright
 * notice and this permission notice appear in all copies of the
 * software, derivative works or modified versions, and any portions
 * thereof, and that both notices appear in supporting documentation.
 *
 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
 *
 * Carnegie Mellon requests users of this software to return to
 *
 *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
 *  School of Computer Science
 *  Carnegie Mellon University
 *  Pittsburgh PA 15213-3890
 *
 * any improvements or extensions that they make and grant Carnegie Mellon
 * the rights to redistribute these changes.
 */
#include <kern/ast.h>
#include <kern/kern_types.h>
#include <kern/mach_param.h>
#include <kern/queue.h>
#include <kern/sched_prim.h>
#include <kern/simple_lock.h>
#include <kern/spl.h>
#include <kern/waitq.h>
#include <kern/zalloc.h>
#include <libkern/OSAtomic.h>
#include <mach/sync_policy.h>
#include <vm/vm_kern.h>

#include <sys/kdebug.h>

#if CONFIG_WAITQ_DEBUG
#define wqdbg(fmt,...) \
        printf("WQ[%s]:  " fmt "\n", __func__, ## __VA_ARGS__)
#else
#define wqdbg(fmt,...) do { } while (0)
#endif

#ifdef WAITQ_VERBOSE_DEBUG
#define wqdbg_v(fmt,...) \
        printf("WQ[v:%s]:  " fmt "\n", __func__, ## __VA_ARGS__)
#else
#define wqdbg_v(fmt,...) do { } while (0)
#endif

#define wqinfo(fmt,...) \
        printf("WQ[%s]: " fmt "\n", __func__,  ## __VA_ARGS__)

#define wqerr(fmt,...) \
        printf("WQ[%s] ERROR: " fmt "\n", __func__, ## __VA_ARGS__)


/*
 * un-comment the following lines to debug the link/prepost tables
 * NOTE: this expands each element by ~40 bytes
 */
//#define CONFIG_WAITQ_LINK_STATS
//#define CONFIG_WAITQ_PREPOST_STATS

/*
 * file-static functions / data
 */
static thread_t waitq_select_one_locked(struct waitq *waitq, event64_t event,
                                        uint64_t *reserved_preposts,
                                        int priority, spl_t *spl);

static kern_return_t waitq_select_thread_locked(struct waitq *waitq,
                                                event64_t event,
                                                thread_t thread, spl_t *spl);

#define WAITQ_SET_MAX (task_max * 3)
static zone_t waitq_set_zone;


#define P2ROUNDUP(x, align) (-(-((uint32_t)(x)) & -(align)))
#define ROUNDDOWN(x,y)  (((x)/(y))*(y))


#ifdef CONFIG_WAITQ_STATS
static __inline__ void waitq_grab_backtrace(uintptr_t bt[NWAITQ_BTFRAMES], int skip);
#endif


/* ----------------------------------------------------------------------
 *
 * Wait Queue Link/Prepost Table Implementation
 *
 * ---------------------------------------------------------------------- */
#define DEFAULT_MIN_FREE_TABLE_ELEM    100
static uint32_t g_min_free_table_elem;
static uint32_t g_min_free_cache;

static vm_size_t   g_wqt_max_tbl_size;
static lck_grp_t   g_wqt_lck_grp;

/* 1 prepost table, 1 setid link table */
#define NUM_WQ_TABLES 2

/* default VA space for waitq tables (zone allocated) */
#define DEFAULT_MAX_TABLE_SIZE  P2ROUNDUP(8 * 1024 * 1024, PAGE_SIZE)

struct wq_id {
        union {
                uint64_t id;
                struct {
                        /*
                         * this bitfied is OK because we don't need to
                         * enforce a particular memory layout
                         */
                        uint64_t idx:18, /* allows indexing up to 8MB of 32byte link objects */
                                 generation:46;
                };
        };
};

enum wqt_elem_type {
        WQT_FREE     = 0,
        WQT_ELEM     = 1,
        WQT_LINK     = 2,
        WQT_RESERVED = 3,
};

struct wqt_elem {
        uint32_t wqt_bits;

        uint32_t wqt_next_idx;

        struct wq_id wqt_id;
};

/* this _must_ match the idx bitfield definition in struct wq_id */
#define WQT_IDX_MAX           (0x3ffff)
#if defined(DEVELOPMENT) || defined(DEBUG)
/* global for lldb macros */
uint64_t g_wqt_idx_max = WQT_IDX_MAX;
#endif

/* reference count bits should _always_ be the low-order bits */
#define WQT_BITS_REFCNT_MASK  (0x1FFFFFFF)
#define WQT_BITS_REFCNT_SHIFT (0)
#define WQT_BITS_REFCNT       (WQT_BITS_REFCNT_MASK << WQT_BITS_REFCNT_SHIFT)

#define WQT_BITS_TYPE_MASK    (0x3)
#define WQT_BITS_TYPE_SHIFT   (29)
#define WQT_BITS_TYPE         (WQT_BITS_TYPE_MASK << WQT_BITS_TYPE_SHIFT)

#define WQT_BITS_VALID_MASK   (0x1)
#define WQT_BITS_VALID_SHIFT  (31)
#define WQT_BITS_VALID        (WQT_BITS_VALID_MASK << WQT_BITS_VALID_SHIFT)

#define wqt_bits_refcnt(bits) \
        (((bits) >> WQT_BITS_REFCNT_SHIFT) & WQT_BITS_REFCNT_MASK)

#define wqt_bits_type(bits) \
        (((bits) >> WQT_BITS_TYPE_SHIFT) & WQT_BITS_TYPE_MASK)

#define wqt_bits_valid(bits) \
        ((bits) & WQT_BITS_VALID)

struct wq_table;
typedef void (*wq_table_poison_func)(struct wq_table *, struct wqt_elem *);

/*
 * A table is a container for slabs of elements. Each slab is 'slab_sz' bytes
 * and contains 'slab_sz/elem_sz' elements (of 'elem_sz' bytes each). These
 * slabs allow the table to be broken up into potentially dis-contiguous VA
 * space. On 32-bit platforms with large amounts of physical RAM, this is
 * quite important. Keeping slabs like this slightly complicates retrieval of
 * table elements, but not by much.
 */
struct wq_table {
        struct wqt_elem **table;   /* an array of 'slabs' of elements */
        struct wqt_elem **next_free_slab;
        struct wq_id     free_list __attribute__((aligned(8)));

        uint32_t         nelem;
        uint32_t         used_elem;
        uint32_t         elem_sz;  /* size of a table element (bytes) */

        uint32_t         slab_sz;  /* size of a table 'slab' object (bytes) */
        uint32_t         slab_shift;
        uint32_t         slab_msk;
        uint32_t         slab_elem;
        zone_t           slab_zone;

        wq_table_poison_func poison;

        lck_mtx_t        lock;
        uint32_t         state;

#if CONFIG_WAITQ_STATS
        uint32_t         nslabs;

        uint64_t         nallocs;
        uint64_t         nreallocs;
        uint64_t         npreposts;
        int64_t          nreservations;
        uint64_t         nreserved_releases;
        uint64_t         nspins;

        uint64_t         max_used;
        uint64_t         avg_used;
        uint64_t         max_reservations;
        uint64_t         avg_reservations;
#endif
} __attribute__((aligned(8)));

#define wqt_elem_ofst_slab(slab, slab_msk, ofst) \
        /* cast through 'void *' to avoid compiler alignment warning messages */ \
        ((struct wqt_elem *)((void *)((uintptr_t)(slab) + ((ofst) & (slab_msk)))))

#if defined(CONFIG_WAITQ_LINK_STATS) || defined(CONFIG_WAITQ_PREPOST_STATS)
/* version that makes no assumption on waste within a slab */
static inline struct wqt_elem *
wqt_elem_idx(struct wq_table *table, uint32_t idx)
{
        int slab_idx = idx / table->slab_elem;
        struct wqt_elem *slab = table->table[slab_idx];
        if (!slab)
                panic("Invalid index:%d slab:%d (NULL) for table:%p\n",
                      idx, slab_idx, table);
        assert(slab->wqt_id.idx <= idx && (slab->wqt_id.idx + table->slab_elem) > idx);
        return wqt_elem_ofst_slab(slab, table->slab_msk, (idx - slab->wqt_id.idx) * table->elem_sz);
}
#else /* !CONFIG_WAITQ_[LINK|PREPOST]_STATS */
/* verion that assumes 100% ultilization of slabs (no waste) */
static inline struct wqt_elem *
wqt_elem_idx(struct wq_table *table, uint32_t idx)
{
        uint32_t ofst = idx * table->elem_sz;
        struct wqt_elem *slab = table->table[ofst >> table->slab_shift];
        if (!slab)
                panic("Invalid index:%d slab:%d (NULL) for table:%p\n",
                      idx, (ofst >> table->slab_shift), table);
        assert(slab->wqt_id.idx <= idx && (slab->wqt_id.idx + table->slab_elem) > idx);
        return wqt_elem_ofst_slab(slab, table->slab_msk, ofst);
}
#endif /* !CONFIG_WAITQ_[LINK|PREPOST]_STATS */

static int __assert_only wqt_elem_in_range(struct wqt_elem *elem,
                                           struct wq_table *table)
{
        struct wqt_elem **base = table->table;
        uintptr_t e = (uintptr_t)elem;
        assert(base != NULL);
        while (*base != NULL) {
                uintptr_t b = (uintptr_t)(*base);
                if (e >= b && e < b + table->slab_sz)
                        return 1;
                base++;
                if ((uintptr_t)base >= (uintptr_t)table->table + PAGE_SIZE)
                        return 0;
        }
        return 0;
}

static struct wqt_elem *wq_table_get_elem(struct wq_table *table, uint64_t id);
static void wq_table_put_elem(struct wq_table *table, struct wqt_elem *elem);
static int wqt_elem_list_link(struct wq_table *table, struct wqt_elem *parent,
                              struct wqt_elem *child);

static void wqt_elem_invalidate(struct wqt_elem *elem)
{
        uint32_t __assert_only old = OSBitAndAtomic(~WQT_BITS_VALID, &elem->wqt_bits);
        OSMemoryBarrier();
        assert(((wqt_bits_type(old) != WQT_RESERVED) && (old & WQT_BITS_VALID)) ||
               ((wqt_bits_type(old) == WQT_RESERVED) && !(old & WQT_BITS_VALID)));
}

static void wqt_elem_mkvalid(struct wqt_elem *elem)
{
        uint32_t __assert_only old = OSBitOrAtomic(WQT_BITS_VALID, &elem->wqt_bits);
        OSMemoryBarrier();
        assert(!(old & WQT_BITS_VALID));
}

static void wqt_elem_set_type(struct wqt_elem *elem, int type)
{
        uint32_t old_bits, new_bits;
        do {
                old_bits = elem->wqt_bits;
                new_bits = (old_bits & ~WQT_BITS_TYPE) |
                           ((type & WQT_BITS_TYPE_MASK) << WQT_BITS_TYPE_SHIFT);
        } while (OSCompareAndSwap(old_bits, new_bits, &elem->wqt_bits) == FALSE);
        OSMemoryBarrier();
}


static void wq_table_bootstrap(void)
{
        uint32_t      tmp32 = 0;

        g_min_free_cache = 0;
        g_min_free_table_elem = DEFAULT_MIN_FREE_TABLE_ELEM;
        if (PE_parse_boot_argn("wqt_min_free", &tmp32, sizeof(tmp32)) == TRUE)
                g_min_free_table_elem = tmp32;
        wqdbg("Minimum free table elements: %d", tmp32);

        g_wqt_max_tbl_size = DEFAULT_MAX_TABLE_SIZE;
        if (PE_parse_boot_argn("wqt_tbl_size", &tmp32, sizeof(tmp32)) == TRUE)
                g_wqt_max_tbl_size = (vm_size_t)P2ROUNDUP(tmp32, PAGE_SIZE);

        lck_grp_init(&g_wqt_lck_grp, "waitq_table_locks", LCK_GRP_ATTR_NULL);
}

static void wq_table_init(struct wq_table *table, const char *name,
                          uint32_t max_tbl_elem, uint32_t elem_sz,
                          wq_table_poison_func poison)
{
        kern_return_t kr;
        uint32_t slab_sz, slab_shift, slab_msk, slab_elem;
        zone_t slab_zone;
        size_t max_tbl_sz;
        struct wqt_elem *e, **base;

        /*
         * First, allocate a single page of memory to act as the base
         * for the table's element slabs
         */
        kr = kernel_memory_allocate(kernel_map, (vm_offset_t *)&base,
                                    PAGE_SIZE, 0, KMA_NOPAGEWAIT, VM_KERN_MEMORY_WAITQ);
        if (kr != KERN_SUCCESS)
                panic("Cannot initialize %s table: "
                      "kernel_memory_allocate failed:%d\n", name, kr);
        memset(base, 0, PAGE_SIZE);

        /*
         * Based on the maximum table size, calculate the slab size:
         * we allocate 1 page of slab pointers for the table, and we need to
         * index elements of 'elem_sz', this gives us the slab size based on
         * the maximum size the table should grow.
         */
        max_tbl_sz = (max_tbl_elem * elem_sz);
        max_tbl_sz = P2ROUNDUP(max_tbl_sz, PAGE_SIZE);

        /* system maximum table size divided by number of slots in a page */
        slab_sz = (uint32_t)(max_tbl_sz / (PAGE_SIZE / (sizeof(void *))));
        if (slab_sz < PAGE_SIZE)
                slab_sz = PAGE_SIZE;

        /* make sure the slab size is a power of two */
        slab_shift = 0;
        slab_msk = ~0;
        for (uint32_t i = 0; i < 31; i++) {
                uint32_t bit = (1 << i);
                if ((slab_sz & bit) == slab_sz) {
                        slab_shift = i;
                        slab_msk = 0;
                        for (uint32_t j = 0; j < i; j++)
                                slab_msk |= (1 << j);
                        break;
                }
                slab_sz &= ~bit;
        }
        slab_elem = slab_sz / elem_sz;

        /* initialize the table's slab zone (for table growth) */
        wqdbg("Initializing %s zone: slab:%d (%d,0x%x) max:%ld",
              name, slab_sz, slab_shift, slab_msk, max_tbl_sz);
        slab_zone = zinit(slab_sz, max_tbl_sz, slab_sz, name);
        assert(slab_zone != ZONE_NULL);

        /* allocate the first slab and populate it */
        base[0] = (struct wqt_elem *)zalloc(slab_zone);
        if (base[0] == NULL)
                panic("Can't allocate a %s table slab from zone:%p",
                      name, slab_zone);

        memset(base[0], 0, slab_sz);

        /* setup the initial freelist */
        wqdbg("initializing %d links (%d bytes each)...", slab_elem, elem_sz);
        for (unsigned l = 0; l < slab_elem; l++) {
                e = wqt_elem_ofst_slab(base[0], slab_msk, l * elem_sz);
                e->wqt_id.idx = l;
                /*
                 * setting generation to 0 ensures that a setid of 0 is
                 * invalid because the generation will be incremented before
                 * each element's allocation.
                 */
                e->wqt_id.generation = 0;
                e->wqt_next_idx = l + 1;
        }

        /* make sure the last free element points to a never-valid idx */
        e = wqt_elem_ofst_slab(base[0], slab_msk, (slab_elem - 1) * elem_sz);
        e->wqt_next_idx = WQT_IDX_MAX;

        lck_mtx_init(&table->lock, &g_wqt_lck_grp, LCK_ATTR_NULL);

        table->slab_sz = slab_sz;
        table->slab_shift = slab_shift;
        table->slab_msk = slab_msk;
        table->slab_elem = slab_elem;
        table->slab_zone = slab_zone;

        table->elem_sz = elem_sz;
        table->nelem = slab_elem;
        table->used_elem = 0;
        table->elem_sz = elem_sz;
        table->poison = poison;

        table->table = base;
        table->next_free_slab = &base[1];
        table->free_list.id = base[0]->wqt_id.id;

#if CONFIG_WAITQ_STATS
        table->nslabs = 1;
        table->nallocs = 0;
        table->nreallocs = 0;
        table->npreposts = 0;
        table->nreservations = 0;
        table->nreserved_releases = 0;

        table->max_used = 0;
        table->avg_used = 0;
        table->max_reservations = 0;
        table->avg_reservations = 0;
#endif
}

/**
 * grow a waitq table by adding another 'slab' of table elements
 *
 * Conditions:
 *      table mutex is unlocked
 *      calling thread can block
 */
static void wq_table_grow(struct wq_table *table, uint32_t min_free)
{
        struct wqt_elem *slab, **slot;
        struct wqt_elem *e = NULL, *first_new_elem, *last_new_elem;
        struct wq_id free_id;
        uint32_t free_elem;

        assert(get_preemption_level() == 0);
        assert(table && table->slab_zone);

        lck_mtx_lock(&table->lock);

        free_elem = table->nelem - table->used_elem;

        /*
         * If the caller just wanted to ensure a minimum number of elements,
         * do that (and don't just blindly grow the table). Also, don't grow
         * the table unnecessarily - we could have been beaten by a higher
         * priority thread who acquired the lock and grew the table before we
         * got here.
         */
        if (free_elem > min_free) {
                lck_mtx_unlock(&table->lock);
                return;
        }

        /* we are now committed to table growth */
        wqdbg_v("BEGIN");

        if (table->next_free_slab == NULL) {
                /*
                 * before we panic, check one more time to see if any other
                 * threads have free'd from space in the table.
                 */
                if ((table->nelem - table->used_elem) > 0) {
                        /* there's at least 1 free element: don't panic yet */
                        lck_mtx_unlock(&table->lock);
                        return;
                }
                panic("No more room to grow table: %p (nelem: %d, used: %d)",
                      table, table->nelem, table->used_elem);
        }
        slot = table->next_free_slab;
        table->next_free_slab++;
        if ((uintptr_t)table->next_free_slab >= (uintptr_t)table->table + PAGE_SIZE)
                table->next_free_slab = NULL;

        assert(*slot == NULL);

        /* allocate another slab */
        slab = (struct wqt_elem *)zalloc(table->slab_zone);
        if (slab == NULL)
                panic("Can't allocate a %s table (%p) slab from zone:%p",
                      table->slab_zone->zone_name, table, table->slab_zone);

        memset(slab, 0, table->slab_sz);

        /* put the new elements into a freelist */
        wqdbg_v("    init %d new links...", table->slab_elem);
        for (unsigned l = 0; l < table->slab_elem; l++) {
                uint32_t idx = l + table->nelem;
                if (idx >= (WQT_IDX_MAX - 1))
                        break; /* the last element of the last slab */
                e = wqt_elem_ofst_slab(slab, table->slab_msk, l * table->elem_sz);
                e->wqt_id.idx = idx;
                e->wqt_next_idx = idx + 1;
        }
        last_new_elem = e;
        assert(last_new_elem != NULL);

        first_new_elem = wqt_elem_ofst_slab(slab, table->slab_msk, 0);

        /* update table book keeping, and atomically swap the freelist head */
        *slot = slab;
        if (table->nelem + table->slab_elem >= WQT_IDX_MAX)
                table->nelem = WQT_IDX_MAX - 1;
        else
                table->nelem += table->slab_elem;

#if CONFIG_WAITQ_STATS
        table->nslabs += 1;
#endif

        /*
         * The atomic swap of the free list head marks the end of table
         * growth. Incoming requests may now use the newly allocated slab
         * of table elements
         */
        free_id = table->free_list;
        /* connect the existing free list to the end of the new free list */
        last_new_elem->wqt_next_idx = free_id.idx;
        while (OSCompareAndSwap64(free_id.id, first_new_elem->wqt_id.id,
                                  &table->free_list.id) == FALSE) {
                OSMemoryBarrier();
                free_id = table->free_list;
                last_new_elem->wqt_next_idx = free_id.idx;
        }
        OSMemoryBarrier();

        lck_mtx_unlock(&table->lock);

        return;
}

static __attribute__((noinline))
struct wqt_elem *wq_table_alloc_elem(struct wq_table *table, int type, int nelem)
{
        int nspins = 0, ntries = 0, nalloc = 0;
        uint32_t table_size;
        struct wqt_elem *elem = NULL;
        struct wq_id free_id, next_id;

        static const int max_retries = 500;

        if (type != WQT_ELEM && type != WQT_LINK && type != WQT_RESERVED)
                panic("wq_table_aloc of invalid elem type:%d from table @%p",
                      type, table);

        assert(nelem > 0);

try_again:
        elem = NULL;
        if (ntries++ > max_retries) {
                struct wqt_elem *tmp;
                if (table->used_elem + nelem >= table_size)
                        panic("No more room to grow table: 0x%p size:%d, used:%d, requested elem:%d",
                              table, table_size, table->used_elem, nelem);
                if (nelem == 1)
                        panic("Too many alloc retries: %d, table:%p, type:%d, nelem:%d",
                              ntries, table, type, nelem);
                /* don't panic: try allocating one-at-a-time */
                while (nelem > 0) {
                        tmp = wq_table_alloc_elem(table, type, 1);
                        if (elem)
                                wqt_elem_list_link(table, tmp, elem);
                        elem = tmp;
                        --nelem;
                }
                assert(elem != NULL);
                return elem;
        }

        nalloc = 0;
        table_size = table->nelem;

        if (table->used_elem + nelem >= table_size) {
                if (get_preemption_level() != 0) {
#if CONFIG_WAITQ_STATS
                        table->nspins += 1;
#endif
                        /*
                         * We may have just raced with table growth: check
                         * again to make sure there really isn't any space.
                         */
                        if (++nspins > 4)
                                panic("Can't grow table %p with preemption"
                                      " disabled!", table);
                        delay(1);
                        goto try_again;
                }
                wq_table_grow(table, nelem);
                goto try_again;
        }

        /* read this value only once before the CAS */
        free_id = table->free_list;
        if (free_id.idx >= table_size)
                goto try_again;

        /*
         * Find the item on the free list which will become the new free list
         * head, but be careful not to modify any memory (read only)!  Other
         * threads can alter table state at any time up until the CAS.  We
         * don't modify any memory until we've successfully swapped out the
         * free list head with the one we've investigated.
         */
        for (struct wqt_elem *next_elem = wqt_elem_idx(table, free_id.idx);
             nalloc < nelem;
             nalloc++) {
                elem = next_elem;
                next_id.generation = 0;
                next_id.idx = next_elem->wqt_next_idx;
                if (next_id.idx < table->nelem) {
                        next_elem = wqt_elem_idx(table, next_id.idx);
                        next_id.id = next_elem->wqt_id.id;
                } else {
                        goto try_again;
                }
        }
        /* 'elem' points to the last element being allocated */

        if (OSCompareAndSwap64(free_id.id, next_id.id,
                               &table->free_list.id) == FALSE)
                goto try_again;

        /* load barrier */
        OSMemoryBarrier();

        /*
         * After the CAS, we know that we own free_id, and it points to a
         * valid table entry (checked above). Grab the table pointer and
         * reset some values.
         */
        OSAddAtomic(nelem, &table->used_elem);

        /* end the list of allocated elements */
        elem->wqt_next_idx = WQT_IDX_MAX;
        /* reset 'elem' to point to the first allocated element */
        elem = wqt_elem_idx(table, free_id.idx);

        /*
         * Update the generation count, and return the element(s)
         * with a single reference (and no valid bit). If the
         * caller immediately calls _put() on any element, then
         * it will be released back to the free list. If the caller
         * subsequently marks the element as valid, then the put
         * will simply drop the reference.
         */
        for (struct wqt_elem *tmp = elem; ; ) {
                assert(!wqt_bits_valid(tmp->wqt_bits) &&
                       (wqt_bits_refcnt(tmp->wqt_bits) == 0));
                --nalloc;
                tmp->wqt_id.generation += 1;
                tmp->wqt_bits = 1;
                wqt_elem_set_type(tmp, type);
                if (tmp->wqt_next_idx == WQT_IDX_MAX)
                        break;
                assert(tmp->wqt_next_idx != WQT_IDX_MAX);
                tmp = wqt_elem_idx(table, tmp->wqt_next_idx);
        }
        assert(nalloc == 0);

#if CONFIG_WAITQ_STATS
        uint64_t nreservations;
        table->nallocs += nelem;
        if (type == WQT_RESERVED)
                OSIncrementAtomic64(&table->nreservations);
        nreservations = table->nreservations;
        if (table->used_elem > table->max_used)
                table->max_used = table->used_elem;
        if (nreservations > table->max_reservations)
                table->max_reservations = nreservations;
        table->avg_used = (table->avg_used + table->used_elem) / 2;
        table->avg_reservations = (table->avg_reservations + nreservations) / 2;
#endif

        return elem;
}

static void wq_table_realloc_elem(struct wq_table *table, struct wqt_elem *elem, int type)
{
        (void)table;
        assert(wqt_elem_in_range(elem, table) &&
               !wqt_bits_valid(elem->wqt_bits));

#if CONFIG_WAITQ_STATS
        table->nreallocs += 1;
        if (wqt_bits_type(elem->wqt_bits) == WQT_RESERVED && type != WQT_RESERVED) {
                /*
                 * This isn't under any lock, so we'll clamp it.
                 * the stats are meant to be informative, not perfectly
                 * accurate
                 */
                OSDecrementAtomic64(&table->nreservations);
        }
        table->avg_reservations = (table->avg_reservations + table->nreservations) / 2;
#endif

        /*
         * Return the same element with a new generation count, and a
         * (potentially) new type. Don't touch the refcount: the caller
         * is responsible for getting that (and the valid bit) correct.
         */
        elem->wqt_id.generation += 1;
        elem->wqt_next_idx = WQT_IDX_MAX;
        wqt_elem_set_type(elem, type);

        return;
}

static void wq_table_free_elem(struct wq_table *table, struct wqt_elem *elem)
{
        struct wq_id next_id;

        assert(wqt_elem_in_range(elem, table) &&
               !wqt_bits_valid(elem->wqt_bits) &&
               (wqt_bits_refcnt(elem->wqt_bits) == 0));

        OSDecrementAtomic(&table->used_elem);

#if CONFIG_WAITQ_STATS
        table->avg_used = (table->avg_used + table->used_elem) / 2;
        if (wqt_bits_type(elem->wqt_bits) == WQT_RESERVED)
                OSDecrementAtomic64(&table->nreservations);
        table->avg_reservations = (table->avg_reservations + table->nreservations) / 2;
#endif

        elem->wqt_bits = 0;

        if (table->poison)
                (table->poison)(table, elem);

again:
        next_id = table->free_list;
        if (next_id.idx >= table->nelem)
                elem->wqt_next_idx = WQT_IDX_MAX;
        else
                elem->wqt_next_idx = next_id.idx;

        /* store barrier */
        OSMemoryBarrier();
        if (OSCompareAndSwap64(next_id.id, elem->wqt_id.id,
                               &table->free_list.id) == FALSE)
                goto again;
}

/* get a reference to a table element identified by 'id' */
static struct wqt_elem *wq_table_get_elem(struct wq_table *table, uint64_t id)
{
        struct wqt_elem *elem;
        uint32_t idx, bits, new_bits;

        /*
         * Here we have a reference to the table which is guaranteed to remain
         * valid until we drop the reference
         */

        idx = ((struct wq_id *)&id)->idx;

        if (idx >= table->nelem)
                panic("id:0x%llx : idx:%d > %d", id, idx, table->nelem);

        elem = wqt_elem_idx(table, idx);

        /* verify the validity by taking a reference on the table object */
        bits = elem->wqt_bits;
        if (!wqt_bits_valid(bits))
                return NULL;

        /*
         * do a pre-verify on the element ID to potentially
         * avoid 2 compare-and-swaps
         */
        if (elem->wqt_id.id != id)
                return NULL;

        new_bits = bits + 1;

        /* check for overflow */
        assert(wqt_bits_refcnt(new_bits) > 0);

        while (OSCompareAndSwap(bits, new_bits, &elem->wqt_bits) == FALSE) {
                /*
                 * either the element became invalid,
                 * or someone else grabbed/removed a reference.
                 */
                bits = elem->wqt_bits;
                if (!wqt_bits_valid(bits)) {
                        /* don't return invalid elements */
                        return NULL;
                }
                new_bits = bits + 1;
                assert(wqt_bits_refcnt(new_bits) > 0);
        }

        /* load barrier */
        OSMemoryBarrier();

        /* check to see that our reference is to the same generation! */
        if (elem->wqt_id.id != id) {
                /*
                wqdbg("ID:0x%llx table generation (%d) != %d",
                      id, elem->wqt_id.generation,
                      ((struct wq_id *)&id)->generation);
                 */
                wq_table_put_elem(table, elem);
                return NULL;
        }

        /* We now have a reference on a valid object */
        return elem;
}

/* release a ref to table element - puts it back on free list as appropriate */
static void wq_table_put_elem(struct wq_table *table, struct wqt_elem *elem)
{
        uint32_t bits, new_bits;

        assert(wqt_elem_in_range(elem, table));

        bits = elem->wqt_bits;
        new_bits = bits - 1;

        /* check for underflow */
        assert(wqt_bits_refcnt(new_bits) < WQT_BITS_REFCNT_MASK);

        while (OSCompareAndSwap(bits, new_bits, &elem->wqt_bits) == FALSE) {
                bits = elem->wqt_bits;
                new_bits = bits - 1;
                /* catch underflow */
                assert(wqt_bits_refcnt(new_bits) < WQT_BITS_REFCNT_MASK);
        }

        /* load barrier */
        OSMemoryBarrier();

        /*
         * if this was the last reference, and it was marked as invalid,
         * then we can add this link object back to the free list
         */
        if (!wqt_bits_valid(new_bits) && (wqt_bits_refcnt(new_bits) == 0))
                wq_table_free_elem(table, elem);

        return;
}


/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 *
 * API: wqt_elem_list_...
 *
 * Reuse the free list linkage member, 'wqt_next_idx' of a table element
 * in a slightly more generic singly-linked list. All members of this
 * list have been allocated from a table, but have not been made valid.
 *
 * - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -*/

/* link parent->child */
static int wqt_elem_list_link(struct wq_table *table, struct wqt_elem *parent, struct wqt_elem *child)
{
        int nelem = 1;

        assert(wqt_elem_in_range(parent, table));

        /* find the end of the parent's list */
        while (parent->wqt_next_idx != WQT_IDX_MAX) {
                assert(parent->wqt_next_idx < table->nelem);
                parent = wqt_elem_idx(table, parent->wqt_next_idx);
                nelem++;
        }

        if (child) {
                assert(wqt_elem_in_range(child, table));
                parent->wqt_next_idx = child->wqt_id.idx;
        }

        return nelem;
}

static struct wqt_elem *wqt_elem_list_next(struct wq_table *table, struct wqt_elem *head)
{
        struct wqt_elem *elem;

        if (!head)
                return NULL;
        if (head->wqt_next_idx >= table->nelem)
                return NULL;

        elem = wqt_elem_idx(table, head->wqt_next_idx);
        assert(wqt_elem_in_range(elem, table));

        return elem;
}

/*
 * Obtain a pointer to the first element of a list.  Don't take an extra
 * reference on the object - the list implicitly holds that reference.
 *
 * This function is used to convert the head of a singly-linked list
 * to a real wqt_elem object.
 */
static struct wqt_elem *wqt_elem_list_first(struct wq_table *table, uint64_t id)
{
        uint32_t idx;
        struct wqt_elem *elem = NULL;

        if (id == 0)
                return NULL;

        idx = ((struct wq_id *)&id)->idx;

        if (idx > table->nelem)
                panic("Invalid element for id:0x%llx", id);
        elem = wqt_elem_idx(table, idx);

        /* invalid element: reserved ID was probably already reallocated */
        if (elem->wqt_id.id != id)
                return NULL;

        /* the returned element should _not_ be marked valid! */
        if (wqt_bits_valid(elem->wqt_bits) ||
            wqt_bits_type(elem->wqt_bits) != WQT_RESERVED ||
            wqt_bits_refcnt(elem->wqt_bits) != 1) {
                panic("Valid/unreserved element %p (0x%x) in reserved list",
                      elem, elem->wqt_bits);
        }

        return elem;
}

static void wqt_elem_reset_next(struct wq_table *table, struct wqt_elem *wqp)
{
        (void)table;

        if (!wqp)
                return;
        assert(wqt_elem_in_range(wqp, table));

        wqp->wqt_next_idx = WQT_IDX_MAX;
}

/*
 * Pop an item off the list.
 * New list head returned in *id, caller responsible for reference on returned
 * object. We do a realloc here to reset the type of the object, but still
 * leave it invalid.
 */
static struct wqt_elem *wqt_elem_list_pop(struct wq_table *table, uint64_t *id, int type)
{
        struct wqt_elem *first, *next;

        if (!id || *id == 0)
                return NULL;

        /* pop an item off the reserved stack */

        first = wqt_elem_list_first(table, *id);
        if (!first) {
                *id = 0;
                return NULL;
        }

        next = wqt_elem_list_next(table, first);
        if (next)
                *id = next->wqt_id.id;
        else
                *id = 0;

        wq_table_realloc_elem(table, first, type);

        return first;
}

/*
 * Free an entire list of linked/reserved elements
 */
static int wqt_elem_list_release(struct wq_table *table,
                                 struct wqt_elem *head,
                                 int __assert_only type)
{
        struct wqt_elem *elem;
        struct wq_id free_id;
        int nelem = 0;

        if (!head)
                return 0;

        for (elem = head; ; ) {
                assert(wqt_elem_in_range(elem, table));
                assert(!wqt_bits_valid(elem->wqt_bits) && (wqt_bits_refcnt(elem->wqt_bits) == 1));
                assert(wqt_bits_type(elem->wqt_bits) == type);

                nelem++;
                elem->wqt_bits = 0;
                if (table->poison)
                        (table->poison)(table, elem);

                if (elem->wqt_next_idx == WQT_IDX_MAX)
                        break;
                assert(elem->wqt_next_idx < table->nelem);
                elem = wqt_elem_idx(table, elem->wqt_next_idx);
        }

        /*
         * 'elem' now points to the end of our list, and 'head' points to the
         * beginning. We want to atomically swap the free list pointer with
         * the 'head' and ensure that 'elem' points to the previous free list
         * head.
         */

again:
        free_id = table->free_list;
        if (free_id.idx >= table->nelem)
                elem->wqt_next_idx = WQT_IDX_MAX;
        else
                elem->wqt_next_idx = free_id.idx;

        /* store barrier */
        OSMemoryBarrier();
        if (OSCompareAndSwap64(free_id.id, head->wqt_id.id,
                               &table->free_list.id) == FALSE)
                goto again;

        OSAddAtomic(-nelem, &table->used_elem);
        return nelem;
}


/* ----------------------------------------------------------------------
 *
 * SetID Link Table Implementation
 *
 * ---------------------------------------------------------------------- */
static struct wq_table g_linktable;

enum setid_link_type {
        SLT_ALL     = -1,
        SLT_FREE    = WQT_FREE,
        SLT_WQS     = WQT_ELEM,
        SLT_LINK    = WQT_LINK,
};

struct setid_link {
        struct wqt_elem wqte;

        union {
                /* wqt_type == SLT_WQS (WQT_ELEM) */
                struct {
                        struct waitq_set *sl_set;
                        /* uint64_t          sl_prepost_id; */
                } sl_wqs;

                /* wqt_type == SLT_LINK (WQT_LINK) */
                struct {
                        uint64_t          sl_left_setid;
                        uint64_t          sl_right_setid;
                } sl_link;
        };
#ifdef CONFIG_WAITQ_LINK_STATS
        thread_t  sl_alloc_th;
        task_t    sl_alloc_task;
        uintptr_t sl_alloc_bt[NWAITQ_BTFRAMES];
        uint64_t  sl_alloc_ts;
        uintptr_t sl_invalidate_bt[NWAITQ_BTFRAMES];
        uint64_t  sl_invalidate_ts;
        uintptr_t sl_mkvalid_bt[NWAITQ_BTFRAMES];
        uint64_t  sl_mkvalid_ts;
        uint64_t  sl_free_ts;
#endif
};
#if !defined(CONFIG_WAITQ_LINK_STATS)
_Static_assert((sizeof(struct setid_link) & (sizeof(struct setid_link) - 1)) == 0,
               "setid_link struct must be a power of two!");
#endif

#define sl_refcnt(link) \
        (wqt_bits_refcnt((link)->wqte.wqt_bits))

#define sl_type(link) \
        (wqt_bits_type((link)->wqte.wqt_bits))

#define sl_set_valid(link) \
        do { \
                wqt_elem_mkvalid(&(link)->wqte); \
                lt_do_mkvalid_stats(&(link)->wqte); \
        } while (0)

#define sl_is_valid(link) \
        wqt_bits_valid((link)->wqte.wqt_bits)

#define sl_set_id wqte.wqt_id

#define SLT_WQS_POISON         ((void *)(0xf00df00d))
#define SLT_LINK_POISON        (0x0bad0badffffffffull)

static void lt_poison(struct wq_table *table, struct wqt_elem *elem)
{
        struct setid_link *sl_link = (struct setid_link *)elem;
        (void)table;

        switch (sl_type(sl_link)) {
        case SLT_WQS:
                sl_link->sl_wqs.sl_set = SLT_WQS_POISON;
                break;
        case SLT_LINK:
                sl_link->sl_link.sl_left_setid = SLT_LINK_POISON;
                sl_link->sl_link.sl_right_setid = SLT_LINK_POISON;
                break;
        default:
                break;
        }
#ifdef CONFIG_WAITQ_LINK_STATS
        memset(sl_link->sl_alloc_bt, 0, sizeof(sl_link->sl_alloc_bt));
        sl_link->sl_alloc_ts = 0;
        memset(sl_link->sl_mkvalid_bt, 0, sizeof(sl_link->sl_mkvalid_bt));
        sl_link->sl_mkvalid_ts = 0;

        sl_link->sl_alloc_th = THREAD_NULL;
        /* leave the sl_alloc_task in place for debugging */

        sl_link->sl_free_ts = mach_absolute_time();
#endif
}

#ifdef CONFIG_WAITQ_LINK_STATS
static __inline__ void lt_do_alloc_stats(struct wqt_elem *elem)
{
        if (elem) {
                struct setid_link *link = (struct setid_link *)elem;
                memset(link->sl_alloc_bt, 0, sizeof(link->sl_alloc_bt));
                waitq_grab_backtrace(link->sl_alloc_bt, 0);
                link->sl_alloc_th = current_thread();
                link->sl_alloc_task = current_task();

                assert(link->sl_alloc_ts == 0);
                link->sl_alloc_ts = mach_absolute_time();

                memset(link->sl_invalidate_bt, 0, sizeof(link->sl_invalidate_bt));
                link->sl_invalidate_ts = 0;
        }
}

static __inline__ void lt_do_invalidate_stats(struct wqt_elem *elem)
{
        struct setid_link *link = (struct setid_link *)elem;

        if (!elem)
                return;

        assert(link->sl_mkvalid_ts > 0);

        memset(link->sl_invalidate_bt, 0, sizeof(link->sl_invalidate_bt));
        link->sl_invalidate_ts = mach_absolute_time();
        waitq_grab_backtrace(link->sl_invalidate_bt, 0);
}

static __inline__ void lt_do_mkvalid_stats(struct wqt_elem *elem)
{
        struct setid_link *link = (struct setid_link *)elem;

        if (!elem)
                return;

        memset(link->sl_mkvalid_bt, 0, sizeof(link->sl_mkvalid_bt));
        link->sl_mkvalid_ts = mach_absolute_time();
        waitq_grab_backtrace(link->sl_mkvalid_bt, 0);
}
#else
#define lt_do_alloc_stats(e)
#define lt_do_invalidate_stats(e)
#define lt_do_mkvalid_stats(e)
#endif /* CONFIG_WAITQ_LINK_STATS */

static void lt_init(void)
{
        uint32_t tablesz = 0, max_links = 0;

        if (PE_parse_boot_argn("wql_tsize", &tablesz, sizeof(tablesz)) != TRUE)
                tablesz = (uint32_t)g_wqt_max_tbl_size;

        tablesz = P2ROUNDUP(tablesz, PAGE_SIZE);
        max_links = tablesz / sizeof(struct setid_link);
        assert(max_links > 0 && tablesz > 0);

        /* we have a restricted index range */
        if (max_links > (WQT_IDX_MAX + 1))
                max_links = WQT_IDX_MAX + 1;

        wqinfo("init linktable with max:%d elements (%d bytes)",
               max_links, tablesz);
        wq_table_init(&g_linktable, "wqslab.links", max_links,
                      sizeof(struct setid_link), lt_poison);
}

static void lt_ensure_free_space(void)
{
        if (g_linktable.nelem - g_linktable.used_elem < g_min_free_table_elem) {
                /*
                 * we don't hold locks on these values, so check for underflow
                 */
                if (g_linktable.used_elem <= g_linktable.nelem) {
                        wqdbg_v("Forcing table growth: nelem=%d, used=%d, min_free=%d",
                                g_linktable.nelem, g_linktable.used_elem,
                                g_min_free_table_elem);
                        wq_table_grow(&g_linktable, g_min_free_table_elem);
                }
        }
}

static struct setid_link *lt_alloc_link(int type)
{
        struct wqt_elem *elem;

        elem = wq_table_alloc_elem(&g_linktable, type, 1);
        lt_do_alloc_stats(elem);
        return (struct setid_link *)elem;
}

static void lt_realloc_link(struct setid_link *link, int type)
{
        wq_table_realloc_elem(&g_linktable, &link->wqte, type);
#ifdef CONFIG_WAITQ_LINK_STATS
        memset(link->sl_alloc_bt, 0, sizeof(link->sl_alloc_bt));
        link->sl_alloc_ts = 0;
        lt_do_alloc_stats(&link->wqte);

        memset(link->sl_invalidate_bt, 0, sizeof(link->sl_invalidate_bt));
        link->sl_invalidate_ts = 0;
#endif
}

static void lt_invalidate(struct setid_link *link)
{
        wqt_elem_invalidate(&link->wqte);
        lt_do_invalidate_stats(&link->wqte);
}

static struct setid_link *lt_get_link(uint64_t setid)
{
        struct wqt_elem *elem;

        elem = wq_table_get_elem(&g_linktable, setid);
        return (struct setid_link *)elem;
}

static void lt_put_link(struct setid_link *link)
{
        if (!link)
                return;
        wq_table_put_elem(&g_linktable, (struct wqt_elem *)link);
}

static struct setid_link *lt_get_reserved(uint64_t setid, int type)
{
        struct wqt_elem *elem;

        elem = wqt_elem_list_first(&g_linktable, setid);
        if (!elem)
                return NULL;
        wq_table_realloc_elem(&g_linktable, elem, type);
        return (struct setid_link *)elem;
}


static inline int waitq_maybe_remove_link(struct waitq *waitq,
                                          uint64_t setid,
                                          struct setid_link *parent,
                                          struct setid_link *left,
                                          struct setid_link *right);

enum {
        LINK_WALK_ONE_LEVEL = 0,
        LINK_WALK_FULL_DAG  = 1,
        LINK_WALK_FULL_DAG_UNLOCKED = 2,
};

typedef int (*lt_callback_func)(struct waitq *waitq, void *ctx,
                                struct setid_link *link);

/**
 * walk all table elements (of type 'link_type') pointed to by 'setid'
 *
 * Conditions:
 *      waitq is locked (or NULL)
 *      'setid' is managed by 'waitq'
 *              this could be direct (waitq->waitq_set_id == setid)
 *              OR indirect (setid is the left/right ID in a LINK chain,
 *                           whose root is waitq->waitq_set_id)
 *
 * Notes:
 *      This function uses recursion to walk the set of table elements
 *      pointed to by 'setid'. For each element encountered, 'cb' will be
 *      called. If non-zero, the return value of this callback function can
 *      early-out of the table walk.
 *
 *      For each link element encountered, the function takes a reference to
 *      it. The reference is dropped only after the callback and any recursion
 *      has completed.
 *
 *      The assumed table/link/tree structure:
 *                   'setid'
 *                   /    \
 *                  /      \
 *              L(LINK)     R(LINK)
 *               /\             /\
 *              /  \           /  \
 *             /    \       Rl(*)  Rr(*)
 *         Ll(*)  Lr(*)      /\    /\
 *           /\     /\    ... ... ... ...
 *        ...  ... ... ...
 *                    \
 *                    WQS(wqset_q.waitq_setid == Sx)
 *                    [waitq set is a membet of setid, 'Sx')
 *
 *                    'Sx'
 *                   /    \
 *                  /      \
 *              L(LINK)     R(LINK)
 *               /\             /\
 *             ... ...        ... ...
 *
 *      The basic algorithm is as follows:
 *      *) take a reference to the table object pointed to by 'setid'
 *      *) if appropriate, call 'cb' (potentially early-out on non-zero return)
 *      *) if the link object points to a waitq set, and the walk type
 *         is 'FULL_DAG' (full directed-acyclic-graph), then try to lock
 *         the associated waitq set object and recursively walk all sets to
 *         which that set belongs. This is a DFS of the tree structure.
 *      *) recurse down the left side of the tree (following the
 *         'sl_left_setid' pointer in the link object
 *      *) recurse down the right side of the tree (following the
 *         'sl_right_setid' pointer in the link object
 */
static __attribute__((noinline))
int walk_setid_links(int walk_type, struct waitq *waitq,
                     uint64_t setid, int link_type,
                     void *ctx, lt_callback_func cb)
{
        struct setid_link *link;
        uint64_t nextid;
        int sl_type;

        link = lt_get_link(setid);

        /* invalid link */
        if (!link)
                return WQ_ITERATE_CONTINUE;

        setid = nextid = 0;
        sl_type = sl_type(link);
        if (sl_type == SLT_LINK) {
                setid  = link->sl_link.sl_left_setid;
                nextid = link->sl_link.sl_right_setid;
        }

        /*
         * Make the callback only on specified link_type (or all links)
         * Note that after the callback, the link object may be
         * invalid. The only valid thing we can do is put our
         * reference to it (which may put it back on the free list)
         */
        if (link_type == SLT_ALL || link_type == sl_type) {
                /* allow the callback to early-out */
                int ret = cb(waitq, ctx, link);
                if (ret != WQ_ITERATE_CONTINUE) {
                        lt_put_link(link);
                        return ret;
                }
        }

        if (sl_type == SLT_WQS &&
            (walk_type == LINK_WALK_FULL_DAG ||
             walk_type == LINK_WALK_FULL_DAG_UNLOCKED)) {
                /*
                 * Recurse down any sets to which this wait queue set was
                 * added.  We do this just before we put our reference to
                 * the link object (which may free it).
                 */
                struct waitq_set *wqset = link->sl_wqs.sl_set;
                int ret = WQ_ITERATE_CONTINUE;
                int get_spl = 0;
                int should_unlock = 0;
                uint64_t wqset_setid = 0;
                spl_t set_spl;

                if (waitq_set_is_valid(wqset) && walk_type == LINK_WALK_FULL_DAG) {
                        if ((!waitq || !waitq_irq_safe(waitq)) &&
                            waitq_irq_safe(&wqset->wqset_q)) {
                                get_spl = 1;
                                set_spl = splsched();
                        }
                        waitq_set_lock(wqset);
                        should_unlock = 1;
                }

                /*
                 * verify the linked waitq set as it could have been
                 * invalidated before we grabbed the lock!
                 */
                if (wqset->wqset_id != link->sl_set_id.id) {
                        /*This is the bottom of the tree: just get out */
                        if (should_unlock) {
                                waitq_set_unlock(wqset);
                                if (get_spl)
                                        splx(set_spl);
                        }
                        lt_put_link(link);
                        return WQ_ITERATE_CONTINUE;
                }

                wqset_setid = wqset->wqset_q.waitq_set_id;

                if (wqset_setid > 0)
                        ret = walk_setid_links(walk_type, &wqset->wqset_q,
                                               wqset_setid, link_type, ctx, cb);
                if (should_unlock) {
                        waitq_set_unlock(wqset);
                        if (get_spl)
                                splx(set_spl);
                }
                if (ret != WQ_ITERATE_CONTINUE) {
                        lt_put_link(link);
                        return ret;
                }
        }

        lt_put_link(link);

        /* recurse down left side of the tree */
        if (setid) {
                int ret = walk_setid_links(walk_type, waitq, setid, link_type, ctx, cb);
                if (ret != WQ_ITERATE_CONTINUE)
                        return ret;
        }

        /* recurse down right side of the tree */
        if (nextid)
                return walk_setid_links(walk_type, waitq, nextid, link_type, ctx, cb);

        return WQ_ITERATE_CONTINUE;
}

/* ----------------------------------------------------------------------
 *
 * Prepost Link Table Implementation
 *
 * ---------------------------------------------------------------------- */
static struct wq_table g_prepost_table;

enum wq_prepost_type {
        WQP_FREE  = WQT_FREE,
        WQP_WQ    = WQT_ELEM,
        WQP_POST  = WQT_LINK,
};

struct wq_prepost {
        struct wqt_elem wqte;

        union {
                /* wqt_type == WQP_WQ (WQT_ELEM) */
                struct {
                        struct waitq *wqp_wq_ptr;
                } wqp_wq;
                /* wqt_type == WQP_POST (WQT_LINK) */
                struct {
                        uint64_t      wqp_next_id;
                        uint64_t      wqp_wq_id;
                } wqp_post;
        };
#ifdef CONFIG_WAITQ_PREPOST_STATS
        thread_t  wqp_alloc_th;
        task_t    wqp_alloc_task;
        uintptr_t wqp_alloc_bt[NWAITQ_BTFRAMES];
#endif
};
#if !defined(CONFIG_WAITQ_PREPOST_STATS)
_Static_assert((sizeof(struct wq_prepost) & (sizeof(struct wq_prepost) - 1)) == 0,
               "wq_prepost struct must be a power of two!");
#endif

#define wqp_refcnt(wqp) \
        (wqt_bits_refcnt((wqp)->wqte.wqt_bits))

#define wqp_type(wqp) \
        (wqt_bits_type((wqp)->wqte.wqt_bits))

#define wqp_set_valid(wqp) \
        wqt_elem_mkvalid(&(wqp)->wqte)

#define wqp_is_valid(wqp) \
        wqt_bits_valid((wqp)->wqte.wqt_bits)

#define wqp_prepostid wqte.wqt_id

#define WQP_WQ_POISON              (0x0bad0badffffffffull)
#define WQP_POST_POISON            (0xf00df00df00df00d)

static void wqp_poison(struct wq_table *table, struct wqt_elem *elem)
{
        struct wq_prepost *wqp = (struct wq_prepost *)elem;
        (void)table;

        switch (wqp_type(wqp)) {
        case WQP_WQ:
                break;
        case WQP_POST:
                wqp->wqp_post.wqp_next_id = WQP_POST_POISON;
                wqp->wqp_post.wqp_wq_id = WQP_POST_POISON;
                break;
        default:
                break;
        }
}

#ifdef CONFIG_WAITQ_PREPOST_STATS
static __inline__ void wqp_do_alloc_stats(struct wqt_elem *elem)
{
        if (elem) {
                struct wq_prepost *wqp = (struct wq_prepost *)elem;

                /* be sure the take stats for _all_ allocated objects */
                for (;;) {
                        uint32_t next_idx;

                        memset(wqp->wqp_alloc_bt, 0, sizeof(wqp->wqp_alloc_bt));
                        waitq_grab_backtrace(wqp->wqp_alloc_bt, 4);
                        wqp->wqp_alloc_th = current_thread();
                        wqp->wqp_alloc_task = current_task();
                        next_idx = wqp->wqte.wqt_next_idx;

                        if (next_idx == WQT_IDX_MAX)
                                break;
                        assert(next_idx < g_prepost_table.nelem);

                        wqp = (struct wq_prepost *)wqt_elem_idx(&g_prepost_table,
                                                                next_idx);
                }
        }
}
#else
#define wqp_do_alloc_stats(e)
#endif /* CONFIG_WAITQ_LINK_STATS */

static void wqp_init(void)
{
        uint32_t tablesz = 0, max_wqp = 0;

        if (PE_parse_boot_argn("wqp_tsize", &tablesz, sizeof(tablesz)) != TRUE)
                tablesz = (uint32_t)g_wqt_max_tbl_size;

        tablesz = P2ROUNDUP(tablesz, PAGE_SIZE);
        max_wqp = tablesz / sizeof(struct wq_prepost);
        assert(max_wqp > 0 && tablesz > 0);

        /* we have a restricted index range */
        if (max_wqp > (WQT_IDX_MAX + 1))
                max_wqp = WQT_IDX_MAX + 1;

        wqinfo("init prepost table with max:%d elements (%d bytes)",
               max_wqp, tablesz);
        wq_table_init(&g_prepost_table, "wqslab.prepost", max_wqp,
                      sizeof(struct wq_prepost), wqp_poison);
}

/*
 * Refill the per-CPU cache.
 */
static void wq_prepost_refill_cpu_cache(uint32_t nalloc)
{
        struct wqt_elem *new_head, *old_head;
        struct wqp_cache *cache;

        /* require preemption enabled to allocate elements */
        if (get_preemption_level() != 0)
                return;

        new_head = wq_table_alloc_elem(&g_prepost_table,
                                       WQT_RESERVED, nalloc);
        if (new_head == NULL)
                return;

        disable_preemption();
        cache = &PROCESSOR_DATA(current_processor(), wqp_cache);
        cache->avail += nalloc;
        if (cache->head == 0 || cache->head == WQT_IDX_MAX) {
                cache->head = new_head->wqt_id.id;
                goto out;
        }

        old_head = wqt_elem_list_first(&g_prepost_table, cache->head);
        (void)wqt_elem_list_link(&g_prepost_table, new_head, old_head);
        cache->head = new_head->wqt_id.id;

out:
        enable_preemption();
        return;
}

static void wq_prepost_ensure_free_space(void)
{
        uint32_t free_elem;
        uint32_t min_free;
        struct wqp_cache *cache;

        if (g_min_free_cache == 0)
                g_min_free_cache = (WQP_CACHE_MAX * ml_get_max_cpus());

        /*
         * Ensure that we always have a pool of per-CPU prepost elements
         */
        disable_preemption();
        cache = &PROCESSOR_DATA(current_processor(), wqp_cache);
        free_elem = cache->avail;
        enable_preemption();

        if (free_elem < (WQP_CACHE_MAX / 3))
                wq_prepost_refill_cpu_cache(WQP_CACHE_MAX - free_elem);

        /*
         * Now ensure that we have a sufficient amount of free table space
         */
        free_elem = g_prepost_table.nelem - g_prepost_table.used_elem;
        min_free = g_min_free_table_elem + g_min_free_cache;
        if (free_elem < min_free) {
                /*
                 * we don't hold locks on these values, so check for underflow
                 */
                if (g_prepost_table.used_elem <= g_prepost_table.nelem) {
                        wqdbg_v("Forcing table growth: nelem=%d, used=%d, min_free=%d+%d",
                                g_prepost_table.nelem, g_prepost_table.used_elem,
                                g_min_free_table_elem, g_min_free_cache);
                        wq_table_grow(&g_prepost_table, min_free);
                }
        }
}

static struct wq_prepost *wq_prepost_alloc(int type, int nelem)
{
        struct wqt_elem *elem;
        struct wq_prepost *wqp;
        struct wqp_cache *cache;

        if (type != WQT_RESERVED)
                goto do_alloc;
        if (nelem == 0)
                return NULL;

        /*
         * First try to grab the elements from the per-CPU cache if we are
         * allocating RESERVED elements
         */
        disable_preemption();
        cache = &PROCESSOR_DATA(current_processor(), wqp_cache);
        if (nelem <= (int)cache->avail) {
                struct wqt_elem *first, *next = NULL;
                int nalloc = nelem;

                cache->avail -= nelem;

                /* grab the first element */
                first = wqt_elem_list_first(&g_prepost_table, cache->head);

                /* find the last element and re-adjust the cache head */
                for (elem = first; elem != NULL && nalloc > 0; elem = next) {
                        next = wqt_elem_list_next(&g_prepost_table, elem);
                        if (--nalloc == 0) {
                                /* terminate the allocated list */
                                elem->wqt_next_idx = WQT_IDX_MAX;
                                break;
                        }
                }
                assert(nalloc == 0);
                if (!next)
                        cache->head = WQT_IDX_MAX;
                else
                        cache->head = next->wqt_id.id;
                /* assert that we don't have mis-matched book keeping */
                assert(!(cache->head == WQT_IDX_MAX && cache->avail > 0));
                enable_preemption();
                elem = first;
                goto out;
        }
        enable_preemption();

do_alloc:
        /* fall-back to standard table allocation */
        elem = wq_table_alloc_elem(&g_prepost_table, type, nelem);
        if (!elem)
                return NULL;

out:
        wqp = (struct wq_prepost *)elem;
        wqp_do_alloc_stats(elem);
        return wqp;
}

/*
static void wq_prepost_realloc(struct wq_prepost *wqp, int type)
{
        wq_table_realloc_elem(&g_prepost_table, &wqp->wqte, type);
}
*/

static void wq_prepost_invalidate(struct wq_prepost *wqp)
{
        wqt_elem_invalidate(&wqp->wqte);
}

static struct wq_prepost *wq_prepost_get(uint64_t wqp_id)
{
        struct wqt_elem *elem;

        elem = wq_table_get_elem(&g_prepost_table, wqp_id);
        return (struct wq_prepost *)elem;
}

static void wq_prepost_put(struct wq_prepost *wqp)
{
        wq_table_put_elem(&g_prepost_table, (struct wqt_elem *)wqp);
}

static int wq_prepost_rlink(struct wq_prepost *parent, struct wq_prepost *child)
{
        return wqt_elem_list_link(&g_prepost_table, &parent->wqte, &child->wqte);
}

static struct wq_prepost *wq_prepost_get_rnext(struct wq_prepost *head)
{
        struct wqt_elem *elem;
        struct wq_prepost *wqp;
        uint64_t id;

        elem = wqt_elem_list_next(&g_prepost_table, &head->wqte);
        if (!elem)
                return NULL;
        id = elem->wqt_id.id;
        elem = wq_table_get_elem(&g_prepost_table, id);

        if (!elem)
                return NULL;
        wqp = (struct wq_prepost *)elem;
        if (elem->wqt_id.id != id ||
            wqp_type(wqp) != WQP_POST ||
            wqp->wqp_post.wqp_next_id != head->wqp_prepostid.id) {
                wq_table_put_elem(&g_prepost_table, elem);
                return NULL;
        }

        return wqp;
}

static void wq_prepost_reset_rnext(struct wq_prepost *wqp)
{
        wqt_elem_reset_next(&g_prepost_table, &wqp->wqte);
}


/**
 * remove 'wqp' from the prepost list on 'wqset'
 *
 * Conditions:
 *      wqset is locked
 *      caller holds a reference on wqp (and is responsible to release it)
 *
 * Result:
 *      wqp is invalidated, wqset is potentially updated with a new
 *      prepost ID, and the next element of the prepost list may be
 *      consumed as well (if the list contained only 2 objects)
 */
static int wq_prepost_remove(struct waitq_set *wqset,
                             struct wq_prepost *wqp)
{
        int more_posts = 1;
        uint64_t next_id = wqp->wqp_post.wqp_next_id;
        uint64_t wqp_id = wqp->wqp_prepostid.id;
        struct wq_prepost *prev_wqp, *next_wqp;

        assert(wqp_type(wqp) == WQP_POST);

        if (next_id == wqp_id) {
                /* the list is singular and becoming empty */
                wqset->wqset_prepost_id = 0;
                more_posts = 0;
                goto out;
        }

        prev_wqp = wq_prepost_get_rnext(wqp);
        assert(prev_wqp != NULL);
        assert(prev_wqp->wqp_post.wqp_next_id == wqp_id);
        assert(prev_wqp->wqp_prepostid.id != wqp_id);
        assert(wqp_type(prev_wqp) == WQP_POST);

        if (prev_wqp->wqp_prepostid.id == next_id) {
                /*
                 * There are two items in the list, and we're removing one. We
                 * only need to keep the WQP_WQ pointer from 'prev_wqp'
                 */
                wqset->wqset_prepost_id = prev_wqp->wqp_post.wqp_wq_id;
                wq_prepost_invalidate(prev_wqp);
                wq_prepost_put(prev_wqp);
                more_posts = 0;
                goto out;
        }

        /* prev->next = next */
        prev_wqp->wqp_post.wqp_next_id = next_id;

        /* next->prev = prev */
        next_wqp = wq_prepost_get(next_id);
        assert(next_wqp != NULL);
        assert(next_wqp != wqp);
        assert(next_wqp != prev_wqp);
        assert(wqp_type(next_wqp) == WQP_POST);

        wq_prepost_reset_rnext(next_wqp);
        wq_prepost_rlink(next_wqp, prev_wqp);

        /* If we remove the head of the list, update the wqset */
        if (wqp_id == wqset->wqset_prepost_id)
                wqset->wqset_prepost_id = next_id;

        wq_prepost_put(prev_wqp);
        wq_prepost_put(next_wqp);

out:
        wq_prepost_reset_rnext(wqp);
        wq_prepost_invalidate(wqp);
        return more_posts;
}

static struct wq_prepost *wq_prepost_rfirst(uint64_t id)
{
        struct wqt_elem *elem;
        elem = wqt_elem_list_first(&g_prepost_table, id);
        wqp_do_alloc_stats(elem);
        return (struct wq_prepost *)(void *)elem;
}

static struct wq_prepost *wq_prepost_rpop(uint64_t *id, int type)
{
        struct wqt_elem *elem;
        elem = wqt_elem_list_pop(&g_prepost_table, id, type);
        wqp_do_alloc_stats(elem);
        return (struct wq_prepost *)(void *)elem;
}

static void wq_prepost_release_rlist(struct wq_prepost *wqp)
{
        int nelem = 0;
        struct wqp_cache *cache;
        struct wqt_elem *elem;

        if (!wqp)
                return;

        elem = &wqp->wqte;

        /*
         * These are reserved elements: release them back to the per-cpu pool
         * if our cache is running low.
         */
        disable_preemption();
        cache = &PROCESSOR_DATA(current_processor(), wqp_cache);
        if (cache->avail < WQP_CACHE_MAX) {
                struct wqt_elem *tmp = NULL;
                if (cache->head != WQT_IDX_MAX)
                        tmp = wqt_elem_list_first(&g_prepost_table, cache->head);
                nelem = wqt_elem_list_link(&g_prepost_table, elem, tmp);
                cache->head = elem->wqt_id.id;
                cache->avail += nelem;
                enable_preemption();
                return;
        }
        enable_preemption();

        /* release these elements back to the main table */
        nelem = wqt_elem_list_release(&g_prepost_table, elem, WQT_RESERVED);

#if CONFIG_WAITQ_STATS
        g_prepost_table.nreserved_releases += 1;
        OSDecrementAtomic64(&g_prepost_table.nreservations);
#endif
}

typedef int (*wqp_callback_func)(struct waitq_set *wqset,
                                 void *ctx,
                                 struct wq_prepost *wqp,
                                 struct waitq *waitq);

/**
 * iterate over a chain of preposts associated with a waitq set.
 *
 * Conditions:
 *      wqset is locked
 *
 * Notes:
 *      This loop performs automatic prepost chain management / culling, and
 *      may reset or adjust the waitq set's prepost ID pointer. If you don't
 *      want this extra processing, you can use wq_prepost_iterate().
 */
static int wq_prepost_foreach_locked(struct waitq_set *wqset,
                                     void *ctx, wqp_callback_func cb)
{
        int ret;
        struct wq_prepost *wqp, *tmp_wqp;

        if (!wqset || !wqset->wqset_prepost_id)
                return WQ_ITERATE_SUCCESS;

restart:
        wqp = wq_prepost_get(wqset->wqset_prepost_id);
        if (!wqp) {
                /*
                 * The prepost object is no longer valid, reset the waitq
                 * set's prepost id.
                 */
                wqset->wqset_prepost_id = 0;
                return WQ_ITERATE_SUCCESS;
        }

        if (wqp_type(wqp) == WQP_WQ) {
                uint64_t __assert_only wqp_id = wqp->wqp_prepostid.id;
                if (cb)
                        ret = cb(wqset, ctx, wqp, wqp->wqp_wq.wqp_wq_ptr);

                switch (ret) {
                case WQ_ITERATE_INVALIDATE_CONTINUE:
                        /* the caller wants to remove the only prepost here */
                        assert(wqp_id == wqset->wqset_prepost_id);
                        wqset->wqset_prepost_id = 0;
                        /* fall through */
                case WQ_ITERATE_CONTINUE:
                        wq_prepost_put(wqp);
                        ret = WQ_ITERATE_SUCCESS;
                        break;
                case WQ_ITERATE_RESTART:
                        wq_prepost_put(wqp);
                        /* fall through */
                case WQ_ITERATE_DROPPED:
                        goto restart;
                default:
                        wq_prepost_put(wqp);
                        break;
                }
                return ret;
        }

        assert(wqp->wqp_prepostid.id == wqset->wqset_prepost_id);
        assert(wqp_type(wqp) == WQP_POST);

        /*
         * At this point we know we have a list of POST objects.
         * Grab a handle to the last element in the list and start
         * the iteration.
         */
        tmp_wqp = wq_prepost_get_rnext(wqp);
        assert(tmp_wqp != NULL && wqp_type(tmp_wqp) == WQP_POST);

        uint64_t last_id = tmp_wqp->wqp_prepostid.id;
        wq_prepost_put(tmp_wqp);

        ret = WQ_ITERATE_SUCCESS;
        for (;;) {
                uint64_t wqp_id, first_id, next_id;

                wqp_id = wqp->wqp_prepostid.id;
                first_id = wqset->wqset_prepost_id;
                next_id = wqp->wqp_post.wqp_next_id;

                /* grab the WQP_WQ object this _POST points to */
                tmp_wqp = wq_prepost_get(wqp->wqp_post.wqp_wq_id);
                if (!tmp_wqp) {
                        /*
                         * This WQP_POST object points to an invalid
                         * WQP_WQ object - remove the POST object from
                         * the list.
                         */
                        if (wq_prepost_remove(wqset, wqp) == 0) {
                                wq_prepost_put(wqp);
                                goto restart;
                        }
                        goto next_prepost;
                }
                assert(wqp_type(tmp_wqp) == WQP_WQ);
                /*
                 * make the callback: note that this could remove 'wqp' or
                 * drop the lock on our waitq set. We need to re-validate
                 * our state when this function returns.
                 */
                if (cb)
                        ret = cb(wqset, ctx, wqp,
                                 tmp_wqp->wqp_wq.wqp_wq_ptr);
                wq_prepost_put(tmp_wqp);

                switch (ret) {
                case WQ_ITERATE_CONTINUE:
                        /* continue iteration */
                        break;
                case WQ_ITERATE_INVALIDATE_CONTINUE:
                        assert(next_id == wqp->wqp_post.wqp_next_id);
                        if (wq_prepost_remove(wqset, wqp) == 0) {
                                wq_prepost_put(wqp);
                                goto restart;
                        }
                        goto next_prepost;
                case WQ_ITERATE_RESTART:
                        wq_prepost_put(wqp);
                        /* fall-through */
                case WQ_ITERATE_DROPPED:
                        /* the callback dropped the ref to wqp: just restart */
                        goto restart;
                default:
                        /* break out of the iteration for some other reason */
                        goto finish_prepost_foreach;
                }

                /*
                 * the set lock may have been dropped during callback,
                 * if something looks different, restart the prepost iteration
                 */
                if (!wqp_is_valid(wqp) ||
                    (wqp->wqp_post.wqp_next_id != next_id) ||
                    wqset->wqset_prepost_id != first_id) {
                        wq_prepost_put(wqp);
                        goto restart;
                }

next_prepost:
                /* this was the last object in the list */
                if (wqp_id == last_id)
                        break;

                /* get the next object */
                tmp_wqp = wq_prepost_get(next_id);
                if (!tmp_wqp) {
                        /*
                         * At this point we've already checked our state
                         * after the callback (which may have dropped the set
                         * lock). If we find an invalid member of the list
                         * then something is wrong.
                         */
                        panic("Invalid WQP_POST member 0x%llx in waitq set "
                              "0x%llx prepost list (first:%llx, "
                              "wqp:%p)",
                              next_id, wqset->wqset_id, first_id, wqp);
                }
                wq_prepost_put(wqp);
                wqp = tmp_wqp;

                assert(wqp_type(wqp) == WQP_POST);
        }

finish_prepost_foreach:
        wq_prepost_put(wqp);
        if (ret == WQ_ITERATE_CONTINUE)
                ret = WQ_ITERATE_SUCCESS;

        return ret;
}

/**
 * Perform a simple loop over a chain of prepost objects
 *
 * Conditions:
 *      If 'prepost_id' is associated with a waitq (set) then that object must
 *      be locked before calling this function.
 *      Callback function, 'cb', must be able to handle a NULL wqset pointer
 *      and a NULL waitq pointer!
 *
 * Notes:
 *      This prepost chain iteration will _not_ automatically adjust any chain
 *      element or linkage. This is the responsibility of the caller! If you
 *      want automatic prepost chain management (at a cost of extra CPU time),
 *      you can use: wq_prepost_foreach_locked().
 */
static int wq_prepost_iterate(uint64_t prepost_id,
                              void *ctx, wqp_callback_func cb)
{
        int ret;
        struct wq_prepost *wqp;

        if (!prepost_id)
                return WQ_ITERATE_SUCCESS;

        wqp = wq_prepost_get(prepost_id);
        if (!wqp)
                return WQ_ITERATE_SUCCESS;

        if (wqp_type(wqp) == WQP_WQ) {
                ret = WQ_ITERATE_SUCCESS;
                if (cb)
                        ret = cb(NULL, ctx, wqp, wqp->wqp_wq.wqp_wq_ptr);

                if (ret != WQ_ITERATE_DROPPED)
                        wq_prepost_put(wqp);
                return ret;
        }

        assert(wqp->wqp_prepostid.id == prepost_id);
        assert(wqp_type(wqp) == WQP_POST);

        /* at this point we know we have a list of POST objects */
        uint64_t next_id;

        ret = WQ_ITERATE_CONTINUE;
        do {
                struct wq_prepost *tmp_wqp;
                struct waitq *wq = NULL;

                next_id = wqp->wqp_post.wqp_next_id;

                /* grab the WQP_WQ object this _POST points to */
                tmp_wqp = wq_prepost_get(wqp->wqp_post.wqp_wq_id);
                if (tmp_wqp) {
                        assert(wqp_type(tmp_wqp) == WQP_WQ);
                        wq = tmp_wqp->wqp_wq.wqp_wq_ptr;
                }

                if (cb)
                        ret = cb(NULL, ctx, wqp, wq);
                if (tmp_wqp)
                        wq_prepost_put(tmp_wqp);

                if (ret != WQ_ITERATE_CONTINUE)
                        break;

                tmp_wqp = wq_prepost_get(next_id);
                if (!tmp_wqp) {
                        /*
                         * the chain is broken: nothing we can do here besides
                         * bail from the iteration.
                         */
                        ret = WQ_ITERATE_ABORTED;
                        break;
                }

                wq_prepost_put(wqp);
                wqp = tmp_wqp;

                assert(wqp_type(wqp) == WQP_POST);
        } while (next_id != prepost_id);

        if (ret != WQ_ITERATE_DROPPED)
                wq_prepost_put(wqp);

        if (ret == WQ_ITERATE_CONTINUE)
                ret = WQ_ITERATE_SUCCESS;
        return ret;
}


struct _is_posted_ctx {
        struct waitq *posting_wq;
        int did_prepost;
};

static int wq_is_preposted_on_set_cb(struct waitq_set *wqset, void *ctx,
                                     struct wq_prepost *wqp, struct waitq *waitq)
{
        struct _is_posted_ctx *pctx = (struct _is_posted_ctx *)ctx;

        (void)wqset;
        (void)wqp;

        /*
         * Don't early-out, run through the _entire_ list:
         * This ensures that we retain a minimum number of invalid elements.
         */
        if (pctx->posting_wq == waitq)
                pctx->did_prepost = 1;

        return WQ_ITERATE_CONTINUE;
}


/**
 * checks if 'waitq' has already preposted on 'wqset'
 *
 * Parameters:
 *      waitq    The waitq that's preposting
 *      wqset    The set onto which waitq may be preposted
 *
 * Conditions:
 *      both waitq and wqset are locked
 *
 * Returns non-zero if 'waitq' has already preposted to 'wqset'
 */
static int wq_is_preposted_on_set(struct waitq *waitq, struct waitq_set *wqset)
{
        int ret;
        struct _is_posted_ctx pctx;

        /*
         * If the set's only prepost matches the waitq's prepost ID,
         * then it obviously already preposted to the set.
         */
        if (waitq->waitq_prepost_id != 0 &&
            wqset->wqset_prepost_id == waitq->waitq_prepost_id)
                return 1;

        /* use full prepost iteration: always trim the list */
        pctx.posting_wq = waitq;
        pctx.did_prepost = 0;
        ret = wq_prepost_foreach_locked(wqset, (void *)&pctx,
                                        wq_is_preposted_on_set_cb);
        return pctx.did_prepost;
}

static struct wq_prepost *wq_get_prepost_obj(uint64_t *reserved, int type)
{
        struct wq_prepost *wqp = NULL;
        /*
         * don't fail just because the caller doesn't have enough
         * reservations, we've kept a low-water mark on the prepost table,
         * so there should be some available for us.
         */
        if (reserved && *reserved) {
                wqp = wq_prepost_rpop(reserved, type);
        } else {
                /*
                 * TODO: if in interrupt context, grab from a special
                 *       region / reserved list!
                 */
                wqp = wq_prepost_alloc(type, 1);
        }

        if (wqp == NULL)
                panic("Couldn't allocate prepost object!");
        return wqp;
}


/**
 * prepost a waitq onto a waitq set
 *
 * Parameters:
 *      wqset    The set onto which waitq will be preposted
 *      waitq    The waitq that's preposting
 *      reserved List (wqt_elem_list_ style) of pre-allocated prepost elements
 *               Could be NULL
 *
 * Conditions:
 *      both wqset and waitq are locked
 *
 * Notes:
 *      If reserved is NULL, this may block on prepost table growth.
 */
static void wq_prepost_do_post_locked(struct waitq_set *wqset,
                                      struct waitq *waitq,
                                      uint64_t *reserved)
{
        struct wq_prepost *wqp_post, *wqp_head, *wqp_tail;

        assert(waitq_held(waitq) && waitq_held(&wqset->wqset_q));

        /*
         * nothing to do if it's already preposted:
         * note that this also culls any invalid prepost objects
         */
        if (wq_is_preposted_on_set(waitq, wqset))
                return;

        /*
         * This function is called because an event is being posted to 'waitq'.
         * We need a prepost object associated with this queue. Allocate one
         * now if the waitq isn't already associated with one.
         */
        if (waitq->waitq_prepost_id == 0) {
                struct wq_prepost *wqp;
                wqp = wq_get_prepost_obj(reserved, WQP_WQ);
                wqp->wqp_wq.wqp_wq_ptr = waitq;
                wqp_set_valid(wqp);
                waitq->waitq_prepost_id = wqp->wqp_prepostid.id;
                wq_prepost_put(wqp);
        }

#if CONFIG_WAITQ_STATS
        g_prepost_table.npreposts += 1;
#endif

        wqdbg_v("preposting waitq %p (0x%llx) to set 0x%llx",
                (void *)VM_KERNEL_UNSLIDE_OR_PERM(waitq),
                waitq->waitq_prepost_id, wqset->wqset_id);

        if (wqset->wqset_prepost_id == 0) {
                /* the set has no previous preposts */
                wqset->wqset_prepost_id = waitq->waitq_prepost_id;
                return;
        }

        wqp_head = wq_prepost_get(wqset->wqset_prepost_id);
        if (!wqp_head) {
                /* the previous prepost has become invalid */
                wqset->wqset_prepost_id = waitq->waitq_prepost_id;
                return;
        }

        assert(wqp_head->wqp_prepostid.id == wqset->wqset_prepost_id);

        /*
         * If we get here, we're going to need at least one new wq_prepost
         * object. If the previous wqset_prepost_id points to a WQP_WQ, we
         * actually need to allocate 2 wq_prepost objects because the WQP_WQ
         * is tied to the waitq and shared across all sets.
         */
        wqp_post = wq_get_prepost_obj(reserved, WQP_POST);

        wqp_post->wqp_post.wqp_wq_id = waitq->waitq_prepost_id;
        wqdbg_v("POST 0x%llx :: WQ 0x%llx", wqp_post->wqp_prepostid.id,
                waitq->waitq_prepost_id);

        if (wqp_type(wqp_head) == WQP_WQ) {
                /*
                 * We must replace the wqset_prepost_id with a pointer
                 * to two new WQP_POST objects
                 */
                uint64_t wqp_id = wqp_head->wqp_prepostid.id;
                wqdbg_v("set 0x%llx previous had 1 WQ prepost (0x%llx): "
                        "replacing with two POST preposts",
                        wqset->wqset_id, wqp_id);

                /* drop the old reference */
                wq_prepost_put(wqp_head);

                /* grab another new object (the 2nd of two) */
                wqp_head = wq_get_prepost_obj(reserved, WQP_POST);

                /* point this one to the original WQP_WQ object */
                wqp_head->wqp_post.wqp_wq_id = wqp_id;
                wqdbg_v("POST 0x%llx :: WQ 0x%llx",
                        wqp_head->wqp_prepostid.id, wqp_id);
        
                /* link it to the new wqp_post object allocated earlier */
                wqp_head->wqp_post.wqp_next_id = wqp_post->wqp_prepostid.id;
                /* make the list a double-linked and circular */
                wq_prepost_rlink(wqp_head, wqp_post);

                /*
                 * Finish setting up the new prepost: point it back to the
                 * POST object we allocated to replace the original wqset
                 * WQ prepost object
                 */
                wqp_post->wqp_post.wqp_next_id = wqp_head->wqp_prepostid.id;
                wq_prepost_rlink(wqp_post, wqp_head);

                /* mark objects valid, and reset the wqset prepost list head */
                wqp_set_valid(wqp_head);
                wqp_set_valid(wqp_post);
                wqset->wqset_prepost_id = wqp_head->wqp_prepostid.id;

                /* release both references */
                wq_prepost_put(wqp_head);
                wq_prepost_put(wqp_post);

                wqdbg_v("set 0x%llx: 0x%llx/0x%llx -> 0x%llx/0x%llx -> 0x%llx",
                        wqset->wqset_id, wqset->wqset_prepost_id,
                        wqp_head->wqp_prepostid.id, wqp_head->wqp_post.wqp_next_id,
                        wqp_post->wqp_prepostid.id,
                        wqp_post->wqp_post.wqp_next_id);
                return;
        }

        assert(wqp_type(wqp_head) == WQP_POST);

        /*
         * Add the new prepost to the end of the prepost list
         */
        wqp_tail = wq_prepost_get_rnext(wqp_head);
        assert(wqp_tail != NULL);
        assert(wqp_tail->wqp_post.wqp_next_id == wqset->wqset_prepost_id);

        /*
         * link the head to the new tail
         * NOTE: this needs to happen first in case wqp_tail == wqp_head
         */
        wq_prepost_reset_rnext(wqp_head);
        wq_prepost_rlink(wqp_head, wqp_post);

        /* point the new object to the list head, and list tail */
        wqp_post->wqp_post.wqp_next_id = wqp_head->wqp_prepostid.id;
        wq_prepost_rlink(wqp_post, wqp_tail);

        /* point the last item in the waitq set's list to the new object */
        wqp_tail->wqp_post.wqp_next_id = wqp_post->wqp_prepostid.id;

        wqp_set_valid(wqp_post);

        wq_prepost_put(wqp_head);
        wq_prepost_put(wqp_tail);
        wq_prepost_put(wqp_post);

        wqdbg_v("set 0x%llx (wqp:0x%llx) last_prepost:0x%llx, "
                "new_prepost:0x%llx->0x%llx", wqset->wqset_id,
                wqset->wqset_prepost_id, wqp_head->wqp_prepostid.id,
                wqp_post->wqp_prepostid.id, wqp_post->wqp_post.wqp_next_id);

        return;
}


/* ----------------------------------------------------------------------
 *
 * Stats collection / reporting
 *
 * ---------------------------------------------------------------------- */
#if CONFIG_WAITQ_STATS
static void wq_table_stats(struct wq_table *table, struct wq_table_stats *stats)
{
        stats->version = WAITQ_STATS_VERSION;
        stats->table_elements = table->nelem;
        stats->table_used_elems = table->used_elem;
        stats->table_elem_sz = table->elem_sz;
        stats->table_slabs = table->nslabs;
        stats->table_slab_sz = table->slab_sz;

        stats->table_num_allocs = table->nallocs;
        stats->table_num_preposts = table->npreposts;
        stats->table_num_reservations = table->nreservations;

        stats->table_max_used = table->max_used;
        stats->table_avg_used = table->avg_used;
        stats->table_max_reservations = table->max_reservations;
        stats->table_avg_reservations = table->avg_reservations;
}

void waitq_link_stats(struct wq_table_stats *stats)
{
        if (!stats)
                return;
        wq_table_stats(&g_linktable, stats);
}

void waitq_prepost_stats(struct wq_table_stats *stats)
{
        wq_table_stats(&g_prepost_table, stats);
}
#endif


/* ----------------------------------------------------------------------
 *
 * Global Wait Queues
 *
 * ---------------------------------------------------------------------- */

static struct waitq g_boot_waitq;
static struct waitq *global_waitqs = &g_boot_waitq;
static uint32_t g_num_waitqs = 1;

/*
 * Zero out the used MSBs of the event.
 */
#define _CAST_TO_EVENT_MASK(event)   ((uintptr_t)(event) & ((1ul << _EVENT_MASK_BITS) - 1ul))

/*
 * The Jenkins "one at a time" hash.
 * TBD: There may be some value to unrolling here,
 * depending on the architecture.
 */
static __inline__ uint32_t waitq_hash(char *key, size_t length)
{
        uint32_t hash = 0;
        size_t i;

        for (i = 0; i < length; i++) {
                hash += key[i];
                hash += (hash << 10);
                hash ^= (hash >> 6);
        }

        hash += (hash << 3);
        hash ^= (hash >> 11);
        hash += (hash << 15);

        hash &= (g_num_waitqs - 1);
        return hash;
}

/* return a global waitq pointer corresponding to the given event */
struct waitq *_global_eventq(char *event, size_t event_length)
{
        return &global_waitqs[waitq_hash(event, event_length)];
}

/* return an indexed global waitq pointer */
struct waitq *global_waitq(int index)
{
        return &global_waitqs[index % g_num_waitqs];
}


#if CONFIG_WAITQ_STATS
/* this global is for lldb */
const uint32_t g_nwaitq_btframes = NWAITQ_BTFRAMES;
struct wq_stats g_boot_stats;
struct wq_stats *g_waitq_stats = &g_boot_stats;

static __inline__ void waitq_grab_backtrace(uintptr_t bt[NWAITQ_BTFRAMES], int skip)
{
        uintptr_t buf[NWAITQ_BTFRAMES + skip];
        if (skip < 0)
                skip = 0;
        memset(buf, 0, (NWAITQ_BTFRAMES + skip) * sizeof(uintptr_t));
        fastbacktrace(buf, g_nwaitq_btframes + skip);
        memcpy(&bt[0], &buf[skip], NWAITQ_BTFRAMES * sizeof(uintptr_t));
}

static __inline__ struct wq_stats *waitq_global_stats(struct waitq *waitq) {
        struct wq_stats *wqs;
        uint32_t idx;

        if (!waitq_is_global(waitq))
                return NULL;

        idx = (uint32_t)(((uintptr_t)waitq - (uintptr_t)global_waitqs) / sizeof(*waitq));
        assert(idx < g_num_waitqs);
        wqs = &g_waitq_stats[idx];
        return wqs;
}

static __inline__ void waitq_stats_count_wait(struct waitq *waitq)
{
        struct wq_stats *wqs = waitq_global_stats(waitq);
        if (wqs != NULL) {
                wqs->waits++;
                waitq_grab_backtrace(wqs->last_wait, 2);
        }
}

static __inline__ void waitq_stats_count_wakeup(struct waitq *waitq)
{
        struct wq_stats *wqs = waitq_global_stats(waitq);
        if (wqs != NULL) {
                wqs->wakeups++;
                waitq_grab_backtrace(wqs->last_wakeup, 2);
        }
}

static __inline__ void waitq_stats_count_clear_wakeup(struct waitq *waitq)
{
        struct wq_stats *wqs = waitq_global_stats(waitq);
        if (wqs != NULL) {
                wqs->wakeups++;
                wqs->clears++;
                waitq_grab_backtrace(wqs->last_wakeup, 2);
        }
}

static __inline__ void waitq_stats_count_fail(struct waitq *waitq)
{
        struct wq_stats *wqs = waitq_global_stats(waitq);
        if (wqs != NULL) {
                wqs->failed_wakeups++;
                waitq_grab_backtrace(wqs->last_failed_wakeup, 2);
        }
}
#else
#define waitq_stats_count_wait(q)         do { } while (0)
#define waitq_stats_count_wakeup(q)       do { } while (0)
#define waitq_stats_count_clear_wakeup(q) do { } while (0)
#define waitq_stats_count_fail(q)         do { } while (0)
#endif

int waitq_is_valid(struct waitq *waitq)
{
        return (waitq != NULL) && ((waitq->waitq_type & ~1) == WQT_QUEUE);
}

int waitq_set_is_valid(struct waitq_set *wqset)
{
        return (wqset != NULL) && waitqs_is_set(wqset);
}

int waitq_is_global(struct waitq *waitq)
{
        if (waitq >= global_waitqs && waitq < global_waitqs + g_num_waitqs)
                return 1;
        return 0;
}

int waitq_irq_safe(struct waitq *waitq)
{
        /* global wait queues have this bit set on initialization */
        return waitq->waitq_irq;
}

static uint32_t waitq_hash_size(void)
{
        uint32_t hsize, queues;
        
        if (PE_parse_boot_argn("wqsize", &hsize, sizeof(hsize)))
                return (hsize);

        queues = thread_max / 11;
        hsize = P2ROUNDUP(queues * sizeof(struct waitq), PAGE_SIZE);

        return hsize;
}

void waitq_bootstrap(void)
{
        kern_return_t kret;
        uint32_t whsize, qsz;

        wq_table_bootstrap();
        lt_init();
        wqp_init();

        /*
         * Determine the amount of memory we're willing to reserve for
         * the waitqueue hash table
         */
        whsize = waitq_hash_size();

        /* Determine the number of waitqueues we can fit. */
        qsz = sizeof(struct waitq);
        whsize = ROUNDDOWN(whsize, qsz);
        g_num_waitqs = whsize / qsz;

        /*
         * The hash algorithm requires that this be a power of 2, so we
         * just mask off all the low-order bits.
         */
        for (uint32_t i = 0; i < 31; i++) {
                uint32_t bit = (1 << i);
                if ((g_num_waitqs & bit) == g_num_waitqs)
                        break;
                g_num_waitqs &= ~bit;
        }
        assert(g_num_waitqs > 0);

        /* Now determine how much memory we really need. */
        whsize = P2ROUNDUP(g_num_waitqs * qsz, PAGE_SIZE);

        wqdbg("allocating %d global queues  (%d bytes)", g_num_waitqs, whsize);
        kret = kernel_memory_allocate(kernel_map, (vm_offset_t *)&global_waitqs,
                                      whsize, 0, KMA_KOBJECT|KMA_NOPAGEWAIT, VM_KERN_MEMORY_WAITQ);
        if (kret != KERN_SUCCESS || global_waitqs == NULL)
                panic("kernel_memory_allocate() failed to alloc global_waitqs"
                      ", error: %d, whsize: 0x%x", kret, whsize);

#if CONFIG_WAITQ_STATS
        whsize = P2ROUNDUP(g_num_waitqs * sizeof(struct wq_stats), PAGE_SIZE);
        kret = kernel_memory_allocate(kernel_map, (vm_offset_t *)&g_waitq_stats,
                                      whsize, 0, KMA_KOBJECT|KMA_NOPAGEWAIT, VM_KERN_MEMORY_WAITQ);
        if (kret != KERN_SUCCESS || global_waitqs == NULL)
                panic("kernel_memory_allocate() failed to alloc g_waitq_stats"
                      ", error: %d, whsize: 0x%x", kret, whsize);
        memset(g_waitq_stats, 0, whsize);
#endif

        for (uint32_t i = 0; i < g_num_waitqs; i++) {
                waitq_init(&global_waitqs[i], SYNC_POLICY_FIFO|SYNC_POLICY_DISABLE_IRQ);
        }


        waitq_set_zone = zinit(sizeof(struct waitq_set),
                               WAITQ_SET_MAX * sizeof(struct waitq_set),
                               sizeof(struct waitq_set),
                               "waitq sets");
        zone_change(waitq_set_zone, Z_NOENCRYPT, TRUE);
}


/* ----------------------------------------------------------------------
 *
 * Wait Queue Implementation
 *
 * ---------------------------------------------------------------------- */

/*
 * Double the standard lock timeout, because wait queues tend
 * to iterate over a number of threads - locking each.  If there is
 * a problem with a thread lock, it normally times out at the wait
 * queue level first, hiding the real problem.
 */
/* For x86, the hardware timeout is in TSC units. */
#if defined(__i386__) || defined(__x86_64__)
#define hwLockTimeOut LockTimeOutTSC
#else
#define hwLockTimeOut LockTimeOut
#endif

void waitq_lock(struct waitq *wq)
{
        if (__improbable(hw_lock_to(&(wq)->waitq_interlock,
                                    hwLockTimeOut * 2) == 0)) {
                boolean_t wql_acquired = FALSE;

                while (machine_timeout_suspended()) {
#if defined(__i386__) || defined(__x86_64__)
                        /*
                         * i386/x86_64 return with preemption disabled on a
                         * timeout for diagnostic purposes.
                         */
                        mp_enable_preemption();
#endif
                        wql_acquired = hw_lock_to(&(wq)->waitq_interlock,
                                                  hwLockTimeOut * 2);
                        if (wql_acquired)
                                break;
                }
                if (wql_acquired == FALSE)
                        panic("waitq deadlock - waitq=%p, cpu=%d\n",
                              wq, cpu_number());
        }
        assert(waitq_held(wq));
}

void waitq_unlock(struct waitq *wq)
{
        assert(waitq_held(wq));
        hw_lock_unlock(&(wq)->waitq_interlock);
}


/**
 * clear the thread-related waitq state
 *
 * Conditions:
 *      'thread' is locked
 */
static inline void thread_clear_waitq_state(thread_t thread)
{
        thread->waitq = NULL;
        thread->wait_event = NO_EVENT64;
        thread->at_safe_point = FALSE;
}


typedef thread_t (*waitq_select_cb)(void *ctx, struct waitq *waitq,
                                    int is_global, thread_t thread);

struct waitq_select_args {
        /* input parameters */
        struct waitq    *posted_waitq;
        struct waitq    *waitq;
        event64_t        event;
        waitq_select_cb  select_cb;
        void            *select_ctx;

        uint64_t        *reserved_preposts;

        /* output parameters */
        queue_t       threadq;
        int           max_threads;
        int          *nthreads;
        spl_t        *spl;
};

static void do_waitq_select_n_locked(struct waitq_select_args *args);

/**
 * callback invoked once for every waitq set to which a waitq belongs
 *
 * Conditions:
 *      ctx->posted_waitq is locked
 *      'link' points to a valid waitq set
 *
 * Notes:
 *      Takes the waitq set lock on the set pointed to by 'link'
 *      Calls do_waitq_select_n_locked() which could recurse back into
 *      this function if the waitq set is a member of other sets.
 *      If no threads were selected, it preposts the input waitq
 *      onto the waitq set pointed to by 'link'.
 */
static int waitq_select_walk_cb(struct waitq *waitq, void *ctx,
                                struct setid_link *link)
{
        int ret = WQ_ITERATE_CONTINUE;
        struct waitq_select_args args = *((struct waitq_select_args *)ctx);
        struct waitq_set *wqset;
        int get_spl = 0;
        spl_t set_spl;

        (void)waitq;
        assert(sl_type(link) == SLT_WQS);

        wqset = link->sl_wqs.sl_set;
        args.waitq = &wqset->wqset_q;

        if (!waitq_irq_safe(waitq) && waitq_irq_safe(&wqset->wqset_q)) {
                get_spl = 1;
                set_spl = splsched();
        }
        waitq_set_lock(wqset);
        /*
         * verify that the link wasn't invalidated just before
         * we were able to take the lock.
         */
        if (wqset->wqset_id != link->sl_set_id.id)
                goto out_unlock;

        /*
         * Find any threads waiting on this wait queue set,
         * and recurse into any waitq set to which this set belongs.
         */
        do_waitq_select_n_locked(&args);

        if (*(args.nthreads) > 0 ||
            (args.threadq && !queue_empty(args.threadq))) {
                /* at least 1 thread was selected and returned: don't prepost */
                if (args.max_threads > 0 &&
                    *(args.nthreads) >= args.max_threads) {
                        /* break out of the setid walk */
                        ret = WQ_ITERATE_FOUND;
                }
                goto out_unlock;
        } else {
                /*
                 * No thread selected: prepost 'waitq' to 'wqset'
                 * if wqset can handle preposts and the event is set to 0.
                 * We also make sure to not post waitq sets to other sets.
                 *
                 * In the future, we may consider an optimization to prepost
                 * 'args.posted_waitq' directly to 'wqset' to avoid
                 * unnecessary data structure manipulations in the kqueue path
                 */
                if (args.event == NO_EVENT64 && waitq_set_can_prepost(wqset)) {
                        wq_prepost_do_post_locked(wqset, waitq,
                                                  args.reserved_preposts);
                }
        }

out_unlock:
        waitq_set_unlock(wqset);
        if (get_spl)
                splx(set_spl);
        return ret;
}

/**
 * generic thread selection from a waitq (and sets to which the waitq belongs)
 *
 * Conditions:
 *      args->waitq (and args->posted_waitq) is locked
 *
 * Notes:
 *      Uses the optional select callback function to refine the selection
 *      of one or more threads from a waitq and any set to which the waitq
 *      belongs. The select callback is invoked once for every thread that
 *      is found to be waiting on the input args->waitq.
 *
 *      If one or more threads are selected, this may disable interrupts.
 *      The previous interrupt state is returned in args->spl and should
 *      be used in a call to splx() if threads are returned to the caller.
 */
static void do_waitq_select_n_locked(struct waitq_select_args *args)
{
        struct waitq *waitq = args->waitq;
        int max_threads = args->max_threads;
        thread_t thread = THREAD_NULL, first_thread = THREAD_NULL;
        int global_q = 0;
        unsigned long eventmask = 0;
        int *nthreads = args->nthreads;

        assert(max_threads != 0);

        global_q = waitq_is_global(waitq);
        if (global_q) {
                eventmask = _CAST_TO_EVENT_MASK(args->event);
                /* make sure this waitq accepts this event mask */
                if ((waitq->waitq_eventmask & eventmask) != eventmask)
                        return;
                eventmask = 0;
        }

        /* look through each thread waiting directly on the waitq */
        qe_foreach_element_safe(thread, &waitq->waitq_queue, links) {
                thread_t t = THREAD_NULL;
                assert(thread->waitq == waitq);
                if (thread->wait_event == args->event) {
                        t = thread;
                        if (first_thread == THREAD_NULL)
                                first_thread = thread;

                        /* allow the caller to futher refine the selection */
                        if (args->select_cb)
                                t = args->select_cb(args->select_ctx, waitq,
                                                    global_q, thread);
                        if (t != THREAD_NULL) {
                                *nthreads += 1;
                                if (args->threadq) {
                                        if (*nthreads == 1)
                                                *(args->spl) = splsched();
                                        thread_lock(t);
                                        thread_clear_waitq_state(t);
                                        /* put locked thread on output queue */
                                        re_queue_tail(args->threadq, &t->links);
                                }
                                /* only enqueue up to 'max' threads */
                                if (*nthreads >= max_threads && max_threads > 0)
                                        break;
                        }
                }
                /* thread wasn't selected, and the waitq is global */
                if (t == THREAD_NULL && global_q)
                        eventmask |= _CAST_TO_EVENT_MASK(thread->wait_event);
        }

        /*
         * Update the eventmask of global queues:
         * - If we selected all the threads in the queue, or we selected zero
         *   threads on the queue, set the eventmask to the calculated value
         *   (potentially 0 if we selected them all)
         * - If we just pulled out a subset of threads from the queue, then we
         *   can't assume the calculated mask is complete (because we may not
         *   have made it through all the threads in the queue), so we have to
         *   leave it alone.
         */
        if (global_q && (queue_empty(&waitq->waitq_queue) || *nthreads == 0))
                waitq->waitq_eventmask = (typeof(waitq->waitq_eventmask))eventmask;

        /*
         * Grab the first thread in the queue if no other thread was selected.
         * We can guarantee that no one has manipulated this thread because
         * it's waiting on the given waitq, and we have that waitq locked.
         */
        if (*nthreads == 0 && first_thread != THREAD_NULL && args->threadq) {
                /* we know this is the first (and only) thread */
                ++(*nthreads);
                *(args->spl) = splsched();
                thread_lock(first_thread);
                thread_clear_waitq_state(first_thread);
                re_queue_tail(args->threadq, &first_thread->links);

                /* update the eventmask on global queues */
                if (global_q && queue_empty(&waitq->waitq_queue))
                        waitq->waitq_eventmask = 0;
        }

        if (max_threads > 0 && *nthreads >= max_threads)
                return;

        /*
         * wait queues that are not in any sets
         * are the bottom of the recursion
         */
        if (!waitq->waitq_set_id)
                return;

        /* check to see if the set ID for this wait queue is valid */
        struct setid_link *link = lt_get_link(waitq->waitq_set_id);
        if (!link) {
                /* the waitq set to which this waitq belonged, has been invalidated */
                waitq->waitq_set_id = 0;
                return;
        }

        lt_put_link(link);

        /*
         * If this waitq is a member of any wait queue sets, we need to look
         * for waiting thread(s) in any of those sets, and prepost all sets that
         * don't have active waiters.
         *
         * Note that we do a local walk of this waitq's links - we manually
         * recurse down wait queue set's with non-zero wqset_q.waitq_set_id
         */
        (void)walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, waitq->waitq_set_id,
                               SLT_WQS, (void *)args, waitq_select_walk_cb);
}

/**
 * main entry point for thread selection from a waitq
 *
 * Conditions:
 *      waitq is locked
 *
 * Returns:
 *      The number of threads waiting on 'waitq' for 'event' which have
 *      been placed onto the input 'threadq'
 *
 * Notes:
 *      The 'select_cb' function is invoked for every thread found waiting
 *      on 'waitq' for 'event'. The thread is _not_ locked upon callback
 *      invocation. This parameter may be NULL.
 *
 *      If one or more threads are returned in 'threadq' then the caller is
 *      responsible to call splx() using the returned 'spl' value. Each
 *      returned thread is locked.
 */
static __inline__ int waitq_select_n_locked(struct waitq *waitq,
                                            event64_t event,
                                            waitq_select_cb select_cb,
                                            void *select_ctx,
                                            uint64_t *reserved_preposts,
                                            queue_t threadq,
                                            int max_threads, spl_t *spl)
{
        int nthreads = 0;

        struct waitq_select_args args = {
                .posted_waitq = waitq,
                .waitq = waitq,
                .event = event,
                .select_cb = select_cb,
                .select_ctx = select_ctx,
                .reserved_preposts = reserved_preposts,
                .threadq = threadq,
                .max_threads = max_threads,
                .nthreads = &nthreads,
                .spl = spl,
        };

        do_waitq_select_n_locked(&args);
        return nthreads;
}


/**
 * callback function that uses thread parameters to determine wakeup eligibility
 *
 * Conditions:
 *      'waitq' is locked
 *      'thread' is not locked
 */
static thread_t waitq_select_one_cb(void *ctx, struct waitq *waitq,
                                    int is_global, thread_t thread)
{
        int fifo_q, realtime;
        boolean_t thread_imp_donor = FALSE;

        (void)ctx;
        (void)waitq;
        (void)is_global;
        realtime = 0;

        fifo_q = 1; /* default to FIFO for all queues for now */
#if IMPORTANCE_INHERITANCE
        if (is_global)
                fifo_q = 0; /* 'thread_imp_donor' takes the place of FIFO checking */
#endif

        if (thread->sched_pri >= BASEPRI_REALTIME)
                realtime = 1;

#if IMPORTANCE_INHERITANCE
        /* 
         * Checking imp donor bit does not need thread lock or
         * or task lock since we have the wait queue lock and
         * thread can not be removed from it without acquiring
         * wait queue lock. The imp donor bit may change
         * once we read its value, but it is ok to wake
         * a thread while someone drops importance assertion
         * on the that thread.
         */
        thread_imp_donor = task_is_importance_donor(thread->task);
#endif /* IMPORTANCE_INHERITANCE */

        if (fifo_q || thread_imp_donor == TRUE
            || realtime || (thread->options & TH_OPT_VMPRIV)) {
                /*
                 * If this thread's task is an importance donor,
                 * or it's a realtime thread, or it's a VM privileged
                 * thread, OR the queue is marked as FIFO:
                 *     select the thread
                 */
                return thread;
        }

        /* by default, _don't_ select the thread */
        return THREAD_NULL;
}

/**
 * select a single thread from a waitq that's waiting for a given event
 *
 * Conditions:
 *      'waitq' is locked
 *
 * Returns:
 *      A locked thread that's been removed from the waitq, but has not
 *      yet been put on a run queue. Caller is responsible to call splx
 *      with the '*spl' value.
 */
static thread_t waitq_select_one_locked(struct waitq *waitq, event64_t event,
                                        uint64_t *reserved_preposts,
                                        int priority, spl_t *spl)
{
        int nthreads;
        queue_head_t threadq;

        (void)priority;

        queue_init(&threadq);

        nthreads = waitq_select_n_locked(waitq, event, waitq_select_one_cb, NULL,
                                         reserved_preposts, &threadq, 1, spl);

        /* if we selected a thread, return it (still locked) */
        if (!queue_empty(&threadq)) {
                thread_t t;
                queue_entry_t qe = dequeue_head(&threadq);
                t = qe_element(qe, struct thread, links);
                assert(queue_empty(&threadq)); /* there should be 1 entry */
                /* t has been locked and removed from all queues */
                return t;
        }

        return THREAD_NULL;
}


struct select_thread_ctx {
        thread_t      thread;
        event64_t     event;
        spl_t        *spl;
};

/**
 * link walk callback invoked once for each set to which a waitq belongs
 *
 * Conditions:
 *      initial waitq is locked
 *      ctx->thread is unlocked
 *
 * Notes:
 *      This may disable interrupts and early-out of the full DAG link walk by
 *      returning KERN_ALREADY_IN_SET. In this case, the returned thread has
 *      been removed from the waitq, it's waitq state has been reset, and the
 *      caller is responsible to call splx() with the returned interrupt state
 *      in ctx->spl.
 */
static int waitq_select_thread_cb(struct waitq *waitq, void *ctx,
                                  struct setid_link *link)
{
        struct select_thread_ctx *stctx = (struct select_thread_ctx *)ctx;
        struct waitq_set *wqset;

        (void)waitq;

        thread_t thread = stctx->thread;
        event64_t event = stctx->event;

        if (sl_type(link) != SLT_WQS)
                return WQ_ITERATE_CONTINUE;

        wqset = link->sl_wqs.sl_set;

        if (!waitq_irq_safe(waitq) && waitq_irq_safe(&wqset->wqset_q)) {
                *(stctx->spl) = splsched();
                waitq_set_lock(wqset);
                thread_lock(thread);
        } else {
                waitq_set_lock(wqset);
                *(stctx->spl) = splsched();
                thread_lock(thread);
        }

        if ((thread->waitq == &wqset->wqset_q)
            && (thread->wait_event == event)) {
                remqueue(&thread->links);
                thread_clear_waitq_state(thread);
                /*
                 * thread still locked,
                 * return non-zero to break out of WQS walk
                 */
                waitq_set_unlock(wqset);
                return WQ_ITERATE_FOUND;
        }

        thread_unlock(thread);
        waitq_set_unlock(wqset);
        splx(*(stctx->spl));

        return WQ_ITERATE_CONTINUE;
}

/**
 * returns KERN_SUCCESS and locks 'thread' if-and-only-if 'thread' is waiting
 * on 'waitq' (or any set to which waitq belongs) for 'event'
 *
 * Conditions:
 *      'waitq' is locked
 *      'thread' is unlocked
 */
static kern_return_t waitq_select_thread_locked(struct waitq *waitq,
                                                event64_t event,
                                                thread_t thread, spl_t *spl)
{
        struct setid_link *link;
        struct select_thread_ctx ctx;
        kern_return_t kr;

        *spl = splsched();
        thread_lock(thread);

        if ((thread->waitq == waitq) && (thread->wait_event == event)) {
                remqueue(&thread->links);
                thread_clear_waitq_state(thread);
                /* thread still locked */
                return KERN_SUCCESS;
        }

        thread_unlock(thread);
        splx(*spl);

        if (!waitq->waitq_set_id)
                return KERN_NOT_WAITING;

        /* check to see if the set ID for this wait queue is valid */
        link = lt_get_link(waitq->waitq_set_id);
        if (!link) {
                /* the waitq to which this set belonged, has been invalidated */
                waitq->waitq_set_id = 0;
                return KERN_NOT_WAITING;
        }

        /*
         * The thread may be waiting on a wait queue set to which
         * the input 'waitq' belongs. Go look for the thread in
         * all wait queue sets. If it's there, we'll remove it
         * because it's equivalent to waiting directly on the input waitq.
         */
        ctx.thread = thread;
        ctx.event = event;
        ctx.spl = spl;
        kr = walk_setid_links(LINK_WALK_FULL_DAG, waitq, waitq->waitq_set_id,
                              SLT_WQS, (void *)&ctx, waitq_select_thread_cb);

        lt_put_link(link);

        /* we found a thread, return success */
        if (kr == WQ_ITERATE_FOUND)
                return KERN_SUCCESS;

        return KERN_NOT_WAITING;
}

static int prepost_exists_cb(struct waitq_set __unused *wqset,
                             void __unused *ctx,
                             struct wq_prepost __unused *wqp,
                             struct waitq __unused *waitq)
{
        /* if we get here, then we know that there is a valid prepost object! */
        return WQ_ITERATE_FOUND;
}

/**
 * declare a thread's intent to wait on 'waitq' for 'wait_event'
 *
 * Conditions:
 *      'waitq' is locked
 *      'thread' is locked
 */
wait_result_t waitq_assert_wait64_locked(struct waitq *waitq,
                                          event64_t wait_event,
                                          wait_interrupt_t interruptible,
                                          wait_timeout_urgency_t urgency,
                                          uint64_t deadline,
                                          uint64_t leeway,
                                          thread_t thread)
{
        wait_result_t wait_result;
        int realtime = 0;

        /*
         * Warning: Do _not_ place debugging print statements here.
         *          The thread is locked!
         */

        if (thread->waitq != NULL)
                panic("thread already waiting on %p", thread->waitq);

        if (waitq_is_set(waitq)) {
                struct waitq_set *wqset = (struct waitq_set *)waitq;
                /*
                 * early-out if the thread is waiting on a wait queue set
                 * that has already been pre-posted.
                 */
                if (wait_event == NO_EVENT64 && waitq_set_maybe_preposted(wqset)) {
                        int ret;
                        /*
                         * Run through the list of potential preposts. Because
                         * this is a hot path, we short-circuit the iteration
                         * if we find just one prepost object.
                         */
                        ret = wq_prepost_foreach_locked(wqset, NULL,
                                                        prepost_exists_cb);
                        if (ret == WQ_ITERATE_FOUND) {
                                thread->wait_result = THREAD_AWAKENED;
                                return THREAD_AWAKENED;
                        }
                }
        }

        /*
         * Realtime threads get priority for wait queue placements.
         * This allows wait_queue_wakeup_one to prefer a waiting
         * realtime thread, similar in principle to performing
         * a wait_queue_wakeup_all and allowing scheduler prioritization
         * to run the realtime thread, but without causing the
         * lock contention of that scenario.
         */
        if (thread->sched_pri >= BASEPRI_REALTIME)
                realtime = 1;

        /*
         * This is the extent to which we currently take scheduling attributes
         * into account.  If the thread is vm priviledged, we stick it at
         * the front of the queue.  Later, these queues will honor the policy
         * value set at waitq_init time.
         */
        wait_result = thread_mark_wait_locked(thread, interruptible);
        /* thread->wait_result has been set */
        if (wait_result == THREAD_WAITING) {
                if (!waitq->waitq_fifo
                    || (thread->options & TH_OPT_VMPRIV) || realtime)
                        enqueue_head(&waitq->waitq_queue, &thread->links);
                else
                        enqueue_tail(&waitq->waitq_queue, &thread->links);

                thread->wait_event = wait_event;
                thread->waitq = waitq;

                if (deadline != 0) {
                        boolean_t act;
                        act = timer_call_enter_with_leeway(&thread->wait_timer,
                                                           NULL,
                                                           deadline, leeway,
                                                           urgency, FALSE);
                        if (!act)
                                thread->wait_timer_active++;
                        thread->wait_timer_is_set = TRUE;
                }

                if (waitq_is_global(waitq))
                        waitq->waitq_eventmask = waitq->waitq_eventmask
                                                | _CAST_TO_EVENT_MASK(wait_event);

                waitq_stats_count_wait(waitq);
        }

        return wait_result;
}

/**
 * remove 'thread' from its current blocking state on 'waitq'
 *
 * Conditions:
 *      'waitq' is locked
 *      'thread' is locked
 *
 * Notes:
 *      This function is primarily used by clear_wait_internal in
 *      sched_prim.c from the thread timer wakeup path
 *      (i.e. the thread was waiting on 'waitq' with a timeout that expired)
 */
void waitq_pull_thread_locked(struct waitq *waitq, thread_t thread)
{
        (void)waitq;
        assert(thread->waitq == waitq);

        remqueue(&thread->links);
        thread_clear_waitq_state(thread);
        waitq_stats_count_clear_wakeup(waitq);

        /* clear the global event mask if this was the last thread there! */
        if (waitq_is_global(waitq) && queue_empty(&waitq->waitq_queue))
                waitq->waitq_eventmask = 0;
}


static __inline__
void maybe_adjust_thread_pri(thread_t thread, int priority) {
        if (thread->sched_pri < priority) {
                if (priority <= MAXPRI) {
                        set_sched_pri(thread, priority);

                        thread->was_promoted_on_wakeup = 1;
                        thread->sched_flags |= TH_SFLAG_PROMOTED;
                }
                return;
        }

        /*
         * If the caller is requesting the waitq subsystem to promote the
         * priority of the awoken thread, then boost the thread's priority to
         * the default WAITQ_BOOST_PRIORITY (if it's not already equal or
         * higher priority).  This boost must be removed via a call to
         * waitq_clear_promotion_locked.
         */
        if (priority == WAITQ_PROMOTE_PRIORITY &&
            (thread->sched_pri < WAITQ_BOOST_PRIORITY ||
             !(thread->sched_flags & TH_SFLAG_WAITQ_PROMOTED))) {

                KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAITQ_PROMOTE) | DBG_FUNC_NONE,
                                      (uintptr_t)thread_tid(thread),
                                      thread->sched_pri, thread->base_pri,
                                      WAITQ_BOOST_PRIORITY, 0);
                thread->sched_flags |= TH_SFLAG_WAITQ_PROMOTED;
                if (thread->sched_pri < WAITQ_BOOST_PRIORITY)
                        set_sched_pri(thread, WAITQ_BOOST_PRIORITY);
        }
}

/**
 * Clear a thread's waitq priority promotion state and the waitq's boost flag
 *
 * This function will always clear the waitq's 'waitq_boost' flag. If the
 * 'thread' parameter is non-null, the this function will also check the
 * priority promotion (boost) state of that thread. If this thread was boosted
 * (by having been awoken from a boosting waitq), then this boost state is
 * cleared. This function is to be paired with waitq_enable_promote_locked.
 */
void waitq_clear_promotion_locked(struct waitq *waitq, thread_t thread)
{
        spl_t s;

        assert(waitq_held(waitq));
        if (thread == THREAD_NULL)
                return;

        if (!waitq_irq_safe(waitq))
                s = splsched();
        thread_lock(thread);

        if (thread->sched_flags & TH_SFLAG_WAITQ_PROMOTED) {
                thread->sched_flags &= ~TH_SFLAG_WAITQ_PROMOTED;

                if (thread->sched_flags & TH_SFLAG_PROMOTED_MASK) {
                        /* it still has other promotions (mutex/rw_lock) */
                } else if (thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) {
                        KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAITQ_DEMOTE) | DBG_FUNC_NONE,
                                              (uintptr_t)thread_tid(thread),
                                              thread->sched_pri,
                                              thread->base_pri,
                                              DEPRESSPRI, 0);
                        set_sched_pri(thread, DEPRESSPRI);
                } else {
                        KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_WAITQ_DEMOTE) | DBG_FUNC_NONE,
                                              (uintptr_t)thread_tid(thread),
                                              thread->sched_pri,
                                              thread->base_pri,
                                              thread->base_pri, 0);
                        thread_recompute_sched_pri(thread, FALSE);
                }
        }

        thread_unlock(thread);
        if (!waitq_irq_safe(waitq))
                splx(s);
}

/**
 * wakeup all threads waiting on 'waitq' for 'wake_event'
 *
 * Conditions:
 *      'waitq' is locked
 *
 * Notes:
 *      May temporarily disable and re-enable interrupts
 *      and re-adjust thread priority of each awoken thread.
 *
 *      If the input 'lock_state' == WAITQ_UNLOCK then the waitq will have
 *      been unlocked before calling thread_go() on any returned threads, and
 *      is guaranteed to be unlocked upon function return.
 */
kern_return_t waitq_wakeup64_all_locked(struct waitq *waitq,
                                        event64_t wake_event,
                                        wait_result_t result,
                                        uint64_t *reserved_preposts,
                                        int priority,
                                        waitq_lock_state_t lock_state)
{
        kern_return_t ret;
        thread_t thread;
        spl_t th_spl;
        int nthreads;
        queue_head_t wakeup_queue;

        assert(waitq_held(waitq));
        queue_init(&wakeup_queue);

        nthreads = waitq_select_n_locked(waitq, wake_event, NULL, NULL,
                                         reserved_preposts,
                                         &wakeup_queue, -1, &th_spl);

        /* set each thread running */
        ret = KERN_NOT_WAITING;

#if CONFIG_WAITQ_STATS
        qe_foreach_element(thread, &wakeup_queue, links)
                waitq_stats_count_wakeup(waitq);
#endif
        if (lock_state == WAITQ_UNLOCK)
                waitq_unlock(waitq);

        qe_foreach_element_safe(thread, &wakeup_queue, links) {
                remqueue(&thread->links);
                maybe_adjust_thread_pri(thread, priority);
                ret = thread_go(thread, result);
                assert(ret == KERN_SUCCESS);
                thread_unlock(thread);
        }
        if (nthreads > 0)
                splx(th_spl);
        else
                waitq_stats_count_fail(waitq);

        return ret;
}

/**
 * wakeup one thread waiting on 'waitq' for 'wake_event'
 *
 * Conditions:
 *      'waitq' is locked
 *
 * Notes:
 *      May temporarily disable and re-enable interrupts.
 */
kern_return_t waitq_wakeup64_one_locked(struct waitq *waitq,
                                        event64_t wake_event,
                                        wait_result_t result,
                                        uint64_t *reserved_preposts,
                                        int priority,
                                        waitq_lock_state_t lock_state)
{
        thread_t thread;
        spl_t th_spl;

        assert(waitq_held(waitq));

        thread = waitq_select_one_locked(waitq, wake_event,
                                         reserved_preposts,
                                         priority, &th_spl);

        if (thread != THREAD_NULL)
                waitq_stats_count_wakeup(waitq);
        else
                waitq_stats_count_fail(waitq);

        if (lock_state == WAITQ_UNLOCK)
                waitq_unlock(waitq);

        if (thread != THREAD_NULL) {
                maybe_adjust_thread_pri(thread, priority);
                kern_return_t ret = thread_go(thread, result);
                assert(ret == KERN_SUCCESS);
                thread_unlock(thread);
                splx(th_spl);
                return ret;
        }

        return KERN_NOT_WAITING;
}

/**
 * wakeup one thread waiting on 'waitq' for 'wake_event'
 *
 * Conditions:
 *      'waitq' is locked
 *
 * Returns:
 *      A locked, runnable thread.
 *      If return value is non-NULL, interrupts have also
 *      been disabled, and the caller is responsible to call
 *      splx() with the returned '*spl' value.
 */
thread_t waitq_wakeup64_identity_locked(struct waitq *waitq,
                                        event64_t wake_event,
                                        wait_result_t result,
                                        spl_t *spl,
                                        uint64_t *reserved_preposts,
                                        waitq_lock_state_t lock_state)
{
        thread_t thread;

        assert(waitq_held(waitq));

        thread = waitq_select_one_locked(waitq, wake_event,
                                         reserved_preposts,
                                         WAITQ_ALL_PRIORITIES, spl);

        if (thread != THREAD_NULL)
                waitq_stats_count_wakeup(waitq);
        else
                waitq_stats_count_fail(waitq);

        if (lock_state == WAITQ_UNLOCK)
                waitq_unlock(waitq);

        if (thread != THREAD_NULL) {
                kern_return_t __assert_only ret;
                ret = thread_go(thread, result);
                assert(ret == KERN_SUCCESS);
        }

        return thread; /* locked if not NULL (caller responsible for spl) */
}

/**
 * wakeup a specific thread iff it's waiting on 'waitq' for 'wake_event'
 *
 * Conditions:
 *      'waitq' is locked
 *      'thread' is unlocked
 *
 * Notes:
 *      May temporarily disable and re-enable interrupts
 *
 *      If the input lock_state == WAITQ_UNLOCK then the waitq will have been
 *      unlocked before calling thread_go() if 'thread' is to be awoken, and
 *      is guaranteed to be unlocked upon function return.
 */
kern_return_t waitq_wakeup64_thread_locked(struct waitq *waitq,
                                           event64_t wake_event,
                                           thread_t thread,
                                           wait_result_t result,
                                           waitq_lock_state_t lock_state)
{
        kern_return_t ret;
        spl_t th_spl;

        assert(waitq_held(waitq));

        /*
         * See if the thread was still waiting there.  If so, it got
         * dequeued and returned locked.
         */
        ret = waitq_select_thread_locked(waitq, wake_event, thread, &th_spl);

        if (ret == KERN_SUCCESS)
                waitq_stats_count_wakeup(waitq);
        else
                waitq_stats_count_fail(waitq);

        if (lock_state == WAITQ_UNLOCK)
                waitq_unlock(waitq);

        if (ret != KERN_SUCCESS)
                return KERN_NOT_WAITING;

        ret = thread_go(thread, result);
        assert(ret == KERN_SUCCESS);
        thread_unlock(thread);
        splx(th_spl);

        return ret;
}



/* ----------------------------------------------------------------------
 *
 * In-Kernel API
 *
 * ---------------------------------------------------------------------- */

/**
 * initialize a waitq object
 */
kern_return_t waitq_init(struct waitq *waitq, int policy)
{
        assert(waitq != NULL);

        /* only FIFO and LIFO for now */
        if ((policy & SYNC_POLICY_FIXED_PRIORITY) != 0)
                return KERN_INVALID_ARGUMENT;

        waitq->waitq_fifo = ((policy & SYNC_POLICY_REVERSED) == 0);
        waitq->waitq_irq = !!(policy & SYNC_POLICY_DISABLE_IRQ);
        waitq->waitq_prepost = 0;
        waitq->waitq_type = WQT_QUEUE;
        waitq->waitq_eventmask = 0;

        waitq->waitq_set_id = 0;
        waitq->waitq_prepost_id = 0;

        hw_lock_init(&waitq->waitq_interlock);
        queue_init(&waitq->waitq_queue);

        return KERN_SUCCESS;
}

struct wq_unlink_ctx {
        struct waitq *unlink_wq;
        struct waitq_set *unlink_wqset;
};

static int waitq_unlink_prepost_cb(struct waitq_set __unused *wqset, void *ctx,
                                   struct wq_prepost *wqp, struct waitq *waitq);

/**
 * walk_setid_links callback to invalidate 'link' parameter
 *
 * Conditions:
 *      Called from walk_setid_links.
 *      Note that unlink other callbacks, this one make no assumptions about
 *      the 'waitq' parameter, specifically it does not have to be locked or
 *      even valid.
 */
static int waitq_unlink_all_cb(struct waitq *waitq, void *ctx,
                               struct setid_link *link)
{
        (void)waitq;
        (void)ctx;
        if (sl_type(link) == SLT_LINK && sl_is_valid(link))
                lt_invalidate(link);

        if (sl_type(link) == SLT_WQS) {
                struct waitq_set *wqset;
                int do_spl = 0;
                spl_t spl;
                struct wq_unlink_ctx ulctx;

                /*
                 * When destroying the waitq, take the time to clear out any
                 * preposts it may have made. This could potentially save time
                 * on the IPC send path which would otherwise have to iterate
                 * over lots of dead port preposts.
                 */
                if (waitq->waitq_prepost_id == 0)
                        goto out;

                wqset = link->sl_wqs.sl_set;
                assert(wqset != NULL);

                if (waitq_set_is_valid(wqset) &&
                    waitq_irq_safe(&wqset->wqset_q)) {
                        spl = splsched();
                        do_spl = 1;
                }
                waitq_set_lock(wqset);

                if (!waitq_set_is_valid(wqset)) {
                        /* someone raced us to teardown */
                        goto out_unlock;
                }
                if (!waitq_set_maybe_preposted(wqset))
                        goto out_unlock;

                ulctx.unlink_wq = waitq;
                ulctx.unlink_wqset = wqset;
                (void)wq_prepost_iterate(wqset->wqset_prepost_id, &ulctx,
                                         waitq_unlink_prepost_cb);
out_unlock:
                waitq_set_unlock(wqset);
                if (do_spl)
                        splx(spl);
        }

out:
        return WQ_ITERATE_CONTINUE;
}


/**
 * cleanup any link/prepost table resources associated with a waitq
 */
void waitq_deinit(struct waitq *waitq)
{
        uint64_t setid = 0;
        spl_t s;

        if (!waitq_valid(waitq))
                return;

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);
        if (!waitq_valid(waitq))
                goto out;

        waitq_unlink_all_locked(waitq, &setid, &s, NULL);
        waitq->waitq_type = WQT_INVALID;
        assert(queue_empty(&waitq->waitq_queue));

out:
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

        if (setid)
                (void)walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, setid,
                                       SLT_ALL, NULL, waitq_unlink_all_cb);
}


/**
 * invalidate the given wq_prepost object
 *
 * Conditions:
 *      Called from wq_prepost_iterate (_not_ from wq_prepost_foreach_locked!)
 */
static int wqset_clear_prepost_chain_cb(struct waitq_set __unused *wqset,
                                        void __unused *ctx,
                                        struct wq_prepost *wqp,
                                        struct waitq __unused *waitq)
{
        if (wqp_type(wqp) == WQP_POST)
                wq_prepost_invalidate(wqp);
        return WQ_ITERATE_CONTINUE;
}


/**
 * allocate and initialize a waitq set object
 *
 * Conditions:
 *      may block
 *
 * Returns:
 *      allocated / initialized waitq_set object
 *      NULL on failure
 */
struct waitq_set *waitq_set_alloc(int policy)
{
        struct waitq_set *wqset;

        wqset = (struct waitq_set *)zalloc(waitq_set_zone);
        if (!wqset)
                panic("Can't allocate a new waitq set from zone %p", waitq_set_zone);

        kern_return_t ret;
        ret = waitq_set_init(wqset, policy, NULL);
        if (ret != KERN_SUCCESS) {
                zfree(waitq_set_zone, wqset);
                wqset = NULL;
        }

        return wqset;
}

/**
 * initialize a waitq set object
 *
 * Conditions:
 *      may (rarely) block if link table needs to grow, and
 *      no 'reserved_link' object is passed.
 */
kern_return_t waitq_set_init(struct waitq_set *wqset,
                             int policy, uint64_t *reserved_link)
{
        struct setid_link *link;
        kern_return_t ret;

        memset(wqset, 0, sizeof(*wqset));

        ret = waitq_init(&wqset->wqset_q, policy);
        if (ret != KERN_SUCCESS)
                return ret;

        wqset->wqset_q.waitq_type = WQT_SET;
        if (policy & SYNC_POLICY_PREPOST)
                wqset->wqset_q.waitq_prepost = 1;
        else
                wqset->wqset_q.waitq_prepost = 0;

        if (reserved_link && *reserved_link != 0) {
                link = lt_get_reserved(*reserved_link, SLT_WQS);
                /* always consume the caller's reference */
                *reserved_link = 0;
        } else {
                link = lt_alloc_link(SLT_WQS);
        }
        if (!link)
                panic("Can't allocate link object for waitq set: %p", wqset);

        link->sl_wqs.sl_set = wqset;
        sl_set_valid(link);

        wqset->wqset_id = link->sl_set_id.id;
        wqset->wqset_prepost_id = 0;
        lt_put_link(link);

        return KERN_SUCCESS;
}

/**
 * clear out / release any resources associated with a waitq set
 *
 * Conditions:
 *      may block
 * Note:
 *      This will render the waitq set invalid, and it must
 *      be re-initialized with waitq_set_init before it can be used again
 */
void waitq_set_deinit(struct waitq_set *wqset)
{
        struct setid_link *link = NULL;
        uint64_t set_id, set_links_id, prepost_id;
        int do_spl = 0;
        spl_t s;

        if (!waitqs_is_set(wqset))
                panic("trying to de-initialize an invalid wqset @%p", wqset);

        if (waitq_irq_safe(&wqset->wqset_q)) {
                s = splsched();
                do_spl = 1;
        }
        waitq_set_lock(wqset);

        set_id = wqset->wqset_id;

        /* grab the set's link object */
        link = lt_get_link(set_id);
        if (link)
                lt_invalidate(link);

        /* someone raced us to deinit */
        if (!link || wqset->wqset_id != set_id || set_id != link->sl_set_id.id) {
                if (link)
                        lt_put_link(link);
                waitq_set_unlock(wqset);
                if (do_spl)
                        splx(s);
                return;
        }

        /* every wait queue set should have a valid link object */
        assert(link != NULL && sl_type(link) == SLT_WQS);

        wqset->wqset_id = 0;

        wqset->wqset_q.waitq_type = WQT_INVALID;
        wqset->wqset_q.waitq_fifo = 0;
        wqset->wqset_q.waitq_prepost = 0;
        /* don't clear the 'waitq_irq' bit: it's used in locking! */
        wqset->wqset_q.waitq_eventmask = 0;

        /*
         * This set may have a lot of preposts, or may have been a member of
         * many other sets. To minimize spinlock hold times, we clear out the
         * waitq set data structure under the lock-hold, but don't clear any
         * table objects. We keep handles to the prepost and set linkage
         * objects and free those outside the critical section.
         */
        prepost_id = wqset->wqset_prepost_id;
        wqset->wqset_prepost_id = 0;

        set_links_id = 0;
        waitq_unlink_all_locked(&wqset->wqset_q, &set_links_id, &s, NULL);

        waitq_set_unlock(wqset);
        if (do_spl)
                splx(s);

        /*
         * walk_setid_links may race with us for access to the waitq set.
         * If walk_setid_links has a reference to the set, then we should wait
         * until the link's refcount goes to 1 (our reference) before we exit
         * this function. That way we ensure that the waitq set memory will
         * remain valid even though it's been cleared out.
         */
        while (sl_refcnt(link) > 1)
                delay(1);
        lt_put_link(link);

        /*
         * release all the set link objects
         * (links to other sets to which this set was previously added)
         */
        if (set_links_id)
                (void)walk_setid_links(LINK_WALK_ONE_LEVEL, NULL, set_links_id,
                                       SLT_ALL, NULL, waitq_unlink_all_cb);

        /* drop / unlink all the prepost table objects */
        (void)wq_prepost_iterate(prepost_id, NULL, wqset_clear_prepost_chain_cb);
}

/**
 * de-initialize and free an allocated waitq set object
 *
 * Conditions:
 *      may block
 */
kern_return_t waitq_set_free(struct waitq_set *wqset)
{
        waitq_set_deinit(wqset);

        memset(wqset, 0, sizeof(*wqset));
        zfree(waitq_set_zone, wqset);

        return KERN_SUCCESS;
}

#if defined(DEVLEOPMENT) || defined(DEBUG)
#if CONFIG_WAITQ_DEBUG
/**
 * return the set ID of 'wqset'
 */
uint64_t wqset_id(struct waitq_set *wqset)
{
        if (!wqset)
                return 0;

        assert(waitqs_is_set(wqset));
        return wqset->wqset_id;
}

/**
 * returns a pointer to the waitq object embedded in 'wqset'
 */
struct waitq *wqset_waitq(struct waitq_set *wqset)
{
        if (!wqset)
                return NULL;

        assert(waitqs_is_set(wqset));

        return &wqset->wqset_q;
}
#endif /* CONFIG_WAITQ_DEBUG */
#endif /* DEVELOPMENT || DEBUG */


/**
 * clear all preposts originating from 'waitq'
 *
 * Conditions:
 *      'waitq' locked
 *      may (rarely) spin waiting for another on-core thread to
 *      release the last reference to the waitq's prepost link object
 *
 * NOTE:
 *      If this function needs to spin, it will drop the waitq lock!
 *      The return value of the function indicates whether or not this
 *      happened: 1 == lock was dropped, 0 == lock held
 */
int waitq_clear_prepost_locked(struct waitq *waitq, spl_t *s)
{
        struct wq_prepost *wqp;
        int dropped_lock = 0;

        if (waitq->waitq_prepost_id == 0)
                return 0;

        wqp = wq_prepost_get(waitq->waitq_prepost_id);
        waitq->waitq_prepost_id = 0;
        if (wqp) {
                uint64_t wqp_id = wqp->wqp_prepostid.id;
                wqdbg_v("invalidate prepost 0x%llx (refcnt:%d)",
                        wqp->wqp_prepostid.id, wqp_refcnt(wqp));
                wq_prepost_invalidate(wqp);
                while (wqp_refcnt(wqp) > 1) {
                        int do_spl = waitq_irq_safe(waitq);

                        /*
                         * Some other thread must have raced us to grab a link
                         * object reference before we invalidated it. This
                         * means that they are probably trying to access the
                         * waitq to which the prepost object points. We need
                         * to wait here until the other thread drops their
                         * reference. We know that no one else can get a
                         * reference (the object has been invalidated), and
                         * that prepost references are short-lived (dropped on
                         * a call to wq_prepost_put). We also know that no one
                         * blocks while holding a reference therefore the
                         * other reference holder must be on-core. We'll just
                         * sit and wait for the other reference to be dropped.
                         */
                        disable_preemption();

                        waitq_unlock(waitq);
                        if (s && do_spl)
                                splx(*s);
                        dropped_lock = 1;
                        /*
                         * don't yield here, just spin and assume the other
                         * consumer is already on core...
                         */
                        delay(1);
                        if (s && do_spl)
                                *s = splsched();
                        waitq_lock(waitq);

                        enable_preemption();
                }
                if (wqp_refcnt(wqp) > 0 && wqp->wqp_prepostid.id == wqp_id)
                        wq_prepost_put(wqp);
        }

        return dropped_lock;
}

/**
 * clear all preposts originating from 'waitq'
 *
 * Conditions:
 *      'waitq' is not locked
 *      may disable and re-enable interrupts
 */
void waitq_clear_prepost(struct waitq *waitq)
{
        spl_t s;
        int do_spl = waitq_irq_safe(waitq);

        assert(waitq_valid(waitq));

        if (do_spl)
                s = splsched();
        waitq_lock(waitq);
        /* it doesn't matter to us if the lock is dropped here */
        (void)waitq_clear_prepost_locked(waitq, &s);
        waitq_unlock(waitq);
        if (do_spl)
                splx(s);
}

/**
 * return a the waitq's prepost object ID (allocate if necessary)
 *
 * Conditions:
 *      'waitq' is unlocked
 */
uint64_t waitq_get_prepost_id(struct waitq *waitq)
{
        struct wq_prepost *wqp;
        uint64_t wqp_id = 0;
        spl_t s;

        if (!waitq_valid(waitq))
                return 0;

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        if (!waitq_valid(waitq))
                goto out_unlock;

        if (waitq->waitq_prepost_id) {
                wqp_id = waitq->waitq_prepost_id;
                goto out_unlock;
        }

        /* don't hold a spinlock while allocating a prepost object */
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

        wqp = wq_prepost_alloc(WQP_WQ, 1);
        if (!wqp)
                return 0;

        /* re-acquire the waitq lock */
        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        if (!waitq_valid(waitq)) {
                wq_prepost_put(wqp);
                wqp_id = 0;
                goto out_unlock;
        }

        if (waitq->waitq_prepost_id) {
                /* we were beat by someone else */
                wq_prepost_put(wqp);
                wqp_id = waitq->waitq_prepost_id;
                goto out_unlock;
        }

        wqp->wqp_wq.wqp_wq_ptr = waitq;

        wqp_set_valid(wqp);
        wqp_id = wqp->wqp_prepostid.id;
        waitq->waitq_prepost_id = wqp_id;

        wq_prepost_put(wqp);

out_unlock:
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

        return wqp_id;
}


static int waitq_inset_cb(struct waitq *waitq, void *ctx, struct setid_link *link)
{
        uint64_t setid = *(uint64_t *)ctx;
        int ltype = sl_type(link);
        (void)waitq;
        if (ltype == SLT_WQS && link->sl_set_id.id == setid) {
                wqdbg_v("  waitq already in set 0x%llx", setid);
                return WQ_ITERATE_FOUND;
        } else if (ltype == SLT_LINK) {
                /*
                 * break out early if we see a link that points to the setid
                 * in question. This saves us a step in the
                 * iteration/recursion
                 */
                wqdbg_v("  waitq already in set 0x%llx (SLT_LINK)", setid);
                if (link->sl_link.sl_left_setid == setid ||
                    link->sl_link.sl_right_setid == setid)
                        return WQ_ITERATE_FOUND;
        }

        return WQ_ITERATE_CONTINUE;
}

/**
 * determine if 'waitq' is a member of 'wqset'
 *
 * Conditions:
 *      neither 'waitq' nor 'wqset' is not locked
 *      may disable and re-enable interrupts while locking 'waitq'
 */
boolean_t waitq_member(struct waitq *waitq, struct waitq_set *wqset)
{
        kern_return_t kr = WQ_ITERATE_SUCCESS;
        uint64_t setid;
        spl_t s;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        if (!waitqs_is_set(wqset))
                return FALSE;

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        setid = wqset->wqset_id;
        if (!setid)
                goto out_unlock;

        /* fast path: most waitqs are members of only 1 set */
        if (waitq->waitq_set_id == setid) {
                waitq_unlock(waitq);
                if (waitq_irq_safe(waitq))
                        splx(s);
                return TRUE;
        }

        /* walk the link table and look for the Set ID of wqset */
        kr = walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, waitq->waitq_set_id,
                              SLT_ALL, (void *)&setid, waitq_inset_cb);

out_unlock:
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

        if (kr == WQ_ITERATE_FOUND)
                return TRUE;
        return FALSE;
}

/**
 * Returns true is the given waitq is a member of at least 1 set
 */
boolean_t waitq_in_set(struct waitq *waitq)
{
        struct setid_link *link;
        boolean_t inset = FALSE;
        spl_t s;

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        if (!waitq->waitq_set_id)
                goto out_unlock;

        link = lt_get_link(waitq->waitq_set_id);
        if (link) {
                /* if we get here, the waitq is in _at_least_one_ set */
                inset = TRUE;
                lt_put_link(link);
        } else {
                /* we can just optimize this for next time */
                waitq->waitq_set_id = 0;
        }

out_unlock:
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);
        return inset;
}


/**
 * pre-allocate a waitq link structure from the link table
 *
 * Conditions:
 *      'waitq' is not locked
 *      may (rarely) block if link table needs to grow
 */
uint64_t waitq_link_reserve(struct waitq *waitq)
{
        struct setid_link *link;
        uint64_t reserved_id = 0;

        assert(get_preemption_level() == 0 && waitq_wait_possible(current_thread()));

        /*
         * We've asserted that the caller can block, so we enforce a
         * minimum-free table element policy here.
         */
        lt_ensure_free_space();

        (void)waitq;
        link = lt_alloc_link(WQT_RESERVED);
        if (!link)
                return 0;

        reserved_id = link->sl_set_id.id;

        return reserved_id;
}

/**
 * release a pre-allocated waitq link structure
 */
void waitq_link_release(uint64_t id)
{
        struct setid_link *link;

        if (id == 0)
                return;

        link = lt_get_reserved(id, SLT_LINK);
        if (!link)
                return;

        /*
         * if we successfully got a link object, then we know
         * it's not been marked valid, and can be released with
         * a standard lt_put_link() which should free the element.
         */
        lt_put_link(link);
#if CONFIG_WAITQ_STATS
        g_linktable.nreserved_releases += 1;
#endif
}

/**
 * link 'waitq' to the set identified by 'setid' using the 'link' structure
 *
 * Conditions:
 *      'waitq' is locked
 *      caller should have a reference to the 'link' object
 */
static kern_return_t waitq_link_internal(struct waitq *waitq,
                                         uint64_t setid, struct setid_link *link)
{
        struct setid_link *qlink;
        kern_return_t kr;

        assert(waitq_held(waitq));

        /*
         * If the waitq_set_id field is empty, then this waitq is not
         * a member of any other set. All we have to do is update the
         * field.
         */
        if (!waitq->waitq_set_id) {
                waitq->waitq_set_id = setid;
                return KERN_SUCCESS;
        }

        qlink = lt_get_link(waitq->waitq_set_id);
        if (!qlink) {
                /*
                 * The set to which this wait queue belonged has been
                 * destroyed / invalidated. We can re-use the waitq field.
                 */
                waitq->waitq_set_id = setid;
                return KERN_SUCCESS;
        }
        lt_put_link(qlink);

        /*
         * Check to see if it's already a member of the set.
         *
         * TODO: check for cycles!
         */
        kr = walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, waitq->waitq_set_id,
                              SLT_ALL, (void *)&setid, waitq_inset_cb);
        if (kr == WQ_ITERATE_FOUND)
                return kr;

        /*
         * This wait queue is a member of at least one set already,
         * and _not_ a member of the given set. Use our previously
         * allocated link object, and hook it up to the wait queue.
         * Note that it's possible that one or more of the wait queue sets to
         * which the wait queue belongs was invalidated before we allocated
         * this link object. That's OK because the next time we use that
         * object we'll just ignore it.
         */
        link->sl_link.sl_left_setid = setid;
        link->sl_link.sl_right_setid = waitq->waitq_set_id;
        sl_set_valid(link);

        waitq->waitq_set_id = link->sl_set_id.id;

        return KERN_SUCCESS;
}

/**
 * link 'waitq' to 'wqset'
 *
 * Conditions:
 *      if 'lock_state' contains WAITQ_SHOULD_LOCK, 'waitq' must be unlocked.
 *      Otherwise, 'waitq' must be locked.
 *
 *      may (rarely) block on link table allocation if the table has to grow,
 *      and no 'reserved_link' object is passed.
 *
 * Notes:
 *      The caller can guarantee that this function will never block by
 *      pre-allocating a link table object and passing its ID in 'reserved_link'
 */
kern_return_t waitq_link(struct waitq *waitq, struct waitq_set *wqset,
                         waitq_lock_state_t lock_state, uint64_t *reserved_link)
{
        kern_return_t kr;
        struct setid_link *link;
        int should_lock = (lock_state == WAITQ_SHOULD_LOCK);
        spl_t s;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        if (!waitqs_is_set(wqset))
                return KERN_INVALID_ARGUMENT;

        wqdbg_v("Link waitq %p to wqset 0x%llx",
                (void *)VM_KERNEL_UNSLIDE_OR_PERM(waitq), wqset->wqset_id);

        if (waitq_irq_safe(waitq) && (!reserved_link || *reserved_link == 0)) {
                /*
                 * wait queues that need IRQs disabled cannot block waiting
                 * for table growth to complete. Even though this is rare,
                 * we require all these waitqs to pass in a reserved link
                 * object to avoid the potential to block.
                 */
                panic("Global/IRQ-safe waitq %p cannot link to %p without"
                      "reserved object!", waitq, wqset);
        }

        /*
         * We _might_ need a new link object here, so we'll grab outside
         * the lock because the alloc call _might_ block.
         *
         * If the caller reserved a link beforehand, then lt_get_link
         * is guaranteed not to block because the caller holds an extra
         * reference to the link which, in turn, hold a reference to the
         * link table.
         */
        if (reserved_link && *reserved_link != 0) {
                link = lt_get_reserved(*reserved_link, SLT_LINK);
                /* always consume the caller's reference */
                *reserved_link = 0;
        } else {
                link = lt_alloc_link(SLT_LINK);
        }
        if (!link)
                return KERN_NO_SPACE;

        if (should_lock) {
                if (waitq_irq_safe(waitq))
                        s = splsched();
                waitq_lock(waitq);
        }

        kr = waitq_link_internal(waitq, wqset->wqset_id, link);

        if (should_lock) {
                waitq_unlock(waitq);
                if (waitq_irq_safe(waitq))
                        splx(s);
        }

        lt_put_link(link);

        return kr;
}

/**
 * helper: unlink 'waitq' from waitq set identified by 'setid'
 *         this function also prunes invalid objects from the tree
 *
 * Conditions:
 *      MUST be called from walk_setid_links link table walk
 *      'waitq' is locked
 *
 * Notes:
 *      This is a helper function which compresses the link table by culling
 *      unused or unnecessary links. See comments below for different
 *      scenarios.
 */
static inline int waitq_maybe_remove_link(struct waitq *waitq,
                                          uint64_t setid,
                                          struct setid_link *parent,
                                          struct setid_link *left,
                                          struct setid_link *right)
{
        uint64_t *wq_setid = &waitq->waitq_set_id;

        /*
         * There are two scenarios:
         *
         * Scenario 1:
         * --------------------------------------------------------------------
         * waitq->waitq_set_id == parent
         *
         *         parent(LINK)
         *           /    \
         *          /      \
         *         /        \
         *  L(LINK/WQS_l)   R(LINK/WQS_r)
         *
         * In this scenario, we assert that the original waitq points to the
         * parent link we were passed in.  If WQS_l (or WQS_r) is the waitq
         * set we're looking for, we can set the corresponding parent
         * link id (left or right) to 0.  To compress the tree, we can reset the
         * waitq_set_id of the original waitq to point to the side of the
         * parent that is still valid. We then discard the parent link object.
         */
        if (*wq_setid == parent->sl_set_id.id) {
                if (!left && !right) {
                        /* completely invalid children */
                        lt_invalidate(parent);
                        wqdbg_v("S1, L+R");
                        *wq_setid = 0;
                        return WQ_ITERATE_INVALID;
                } else if (!left || left->sl_set_id.id == setid) {
                        /*
                         * left side matches we know it points either to the
                         * WQS we're unlinking, or to an invalid object:
                         * no need to invalidate it
                         */
                        *wq_setid = right ? right->sl_set_id.id : 0;
                        lt_invalidate(parent);
                        wqdbg_v("S1, L");
                        return left ? WQ_ITERATE_UNLINKED : WQ_ITERATE_INVALID;
                } else if (!right || right->sl_set_id.id == setid) {
                        /*
                         * if right side matches we know it points either to the
                         * WQS we're unlinking, or to an invalid object:
                         * no need to invalidate it
                         */
                        *wq_setid = left ? left->sl_set_id.id : 0;
                        lt_invalidate(parent);
                        wqdbg_v("S1, R");
                        return right ? WQ_ITERATE_UNLINKED : WQ_ITERATE_INVALID;
                }
        }

        /*
         * the tree walk starts at the top-of-tree and moves down,
         * so these are safe asserts.
         */
        assert(left || right); /* one of them has to be valid at this point */

        /*
         * Scenario 2:
         * --------------------------------------------------------------------
         * waitq->waitq_set_id == ... (OR parent)
         *
         *                    ...
         *                     |
         *                   parent
         *                   /    \
         *                  /      \
         *              L(LINK)     R(LINK)
         *               /\             /\
         *              /  \           /  \
         *             /    \       Rl(*)  Rr(*)
         *         Ll(WQS)  Lr(WQS)
         *
         * In this scenario, a leaf node of either the left or right side
         * could be the wait queue set we're looking to unlink. We also handle
         * the case where one of these links is invalid.  If a leaf node is
         * invalid or it's the set we're looking for, we can safely remove the
         * middle link (left or right) and point the parent link directly to
         * the remaining leaf node.
         */
        if (left && sl_type(left) == SLT_LINK) {
                uint64_t Ll, Lr;
                struct setid_link *linkLl, *linkLr;
                assert(left->sl_set_id.id != setid);
                Ll = left->sl_link.sl_left_setid;
                Lr = left->sl_link.sl_right_setid;
                linkLl = lt_get_link(Ll);
                linkLr = lt_get_link(Lr);
                if (!linkLl && !linkLr) {
                        /*
                         * The left object points to two invalid objects!
                         * We can invalidate the left w/o touching the parent.
                         */
                        lt_invalidate(left);
                        wqdbg_v("S2, Ll+Lr");
                        return WQ_ITERATE_INVALID;
                } else if (!linkLl || Ll == setid) {
                        /* Ll is invalid and/or the wait queue set we're looking for */
                        parent->sl_link.sl_left_setid = Lr;
                        lt_invalidate(left);
                        lt_put_link(linkLl);
                        lt_put_link(linkLr);
                        wqdbg_v("S2, Ll");
                        return linkLl ? WQ_ITERATE_UNLINKED : WQ_ITERATE_INVALID;
                } else if (!linkLr || Lr == setid) {
                        /* Lr is invalid and/or the wait queue set we're looking for */
                        parent->sl_link.sl_left_setid = Ll;
                        lt_invalidate(left);
                        lt_put_link(linkLr);
                        lt_put_link(linkLl);
                        wqdbg_v("S2, Lr");
                        return linkLr ? WQ_ITERATE_UNLINKED : WQ_ITERATE_INVALID;
                }
                lt_put_link(linkLl);
                lt_put_link(linkLr);
        }

        if (right && sl_type(right) == SLT_LINK) {
                uint64_t Rl, Rr;
                struct setid_link *linkRl, *linkRr;
                assert(right->sl_set_id.id != setid);
                Rl = right->sl_link.sl_left_setid;
                Rr = right->sl_link.sl_right_setid;
                linkRl = lt_get_link(Rl);
                linkRr = lt_get_link(Rr);
                if (!linkRl && !linkRr) {
                        /*
                         * The right object points to two invalid objects!
                         * We can invalidate the right w/o touching the parent.
                         */
                        lt_invalidate(right);
                        wqdbg_v("S2, Rl+Rr");
                        return WQ_ITERATE_INVALID;
                } else if (!linkRl || Rl == setid) {
                        /* Rl is invalid and/or the wait queue set we're looking for */
                        parent->sl_link.sl_right_setid = Rr;
                        lt_invalidate(right);
                        lt_put_link(linkRl);
                        lt_put_link(linkRr);
                        wqdbg_v("S2, Rl");
                        return linkRl ? WQ_ITERATE_UNLINKED : WQ_ITERATE_INVALID;
                } else if (!linkRr || Rr == setid) {
                        /* Rr is invalid and/or the wait queue set we're looking for */
                        parent->sl_link.sl_right_setid = Rl;
                        lt_invalidate(right);
                        lt_put_link(linkRl);
                        lt_put_link(linkRr);
                        wqdbg_v("S2, Rr");
                        return linkRr ? WQ_ITERATE_UNLINKED : WQ_ITERATE_INVALID;
                }
                lt_put_link(linkRl);
                lt_put_link(linkRr);
        }

        return WQ_ITERATE_CONTINUE;
}

/**
 * link table walk callback that unlinks 'waitq' from 'ctx->setid'
 *
 * Conditions:
 *      called from walk_setid_links
 *      'waitq' is locked
 *
 * Notes:
 *      uses waitq_maybe_remove_link() to compress the linktable and
 *      perform the actual unlinking
 */
static int waitq_unlink_cb(struct waitq *waitq, void *ctx,
                           struct setid_link *link)
{
        uint64_t setid = *((uint64_t *)ctx);
        struct setid_link *right, *left;
        int ret = 0;

        if (sl_type(link) != SLT_LINK)
                return WQ_ITERATE_CONTINUE;

        do  {
                left  = lt_get_link(link->sl_link.sl_left_setid);
                right = lt_get_link(link->sl_link.sl_right_setid);

                ret = waitq_maybe_remove_link(waitq, setid, link, left, right);

                lt_put_link(left);
                lt_put_link(right);

                if (!sl_is_valid(link))
                        return WQ_ITERATE_INVALID;
                /* A ret value of UNLINKED will break us out of table walk */
        } while (ret == WQ_ITERATE_INVALID);

        return ret;
}


/**
 * undo/remove a prepost from 'ctx' (waitq) to 'wqset'
 *
 * Conditions:
 *      Called from wq_prepost_foreach_locked OR wq_prepost_iterate
 *      'wqset' may be NULL
 *      (ctx)->unlink_wqset is locked
 */
static int waitq_unlink_prepost_cb(struct waitq_set __unused *wqset, void *ctx,
                                   struct wq_prepost *wqp, struct waitq *waitq)
{
        struct wq_unlink_ctx *ulctx = (struct wq_unlink_ctx *)ctx;

        if (waitq != ulctx->unlink_wq)
                return WQ_ITERATE_CONTINUE;

        if (wqp_type(wqp) == WQP_WQ &&
            wqp->wqp_prepostid.id == ulctx->unlink_wqset->wqset_prepost_id) {
                /* this is the only prepost on this wait queue set */
                wqdbg_v("unlink wqp (WQ) 0x%llx", wqp->wqp_prepostid.id);
                ulctx->unlink_wqset->wqset_prepost_id = 0;
                return WQ_ITERATE_BREAK;
        }

        assert(wqp_type(wqp) == WQP_POST);

        /*
         * The prepost object 'wqp' points to a waitq which should no longer
         * be preposted to 'ulctx->unlink_wqset'. We can remove the prepost
         * object from the list and break out of the iteration. Using the
         * context object in this way allows this same callback function to be
         * used from both wq_prepost_foreach_locked and wq_prepost_iterate.
         */
        wq_prepost_remove(ulctx->unlink_wqset, wqp);
        return WQ_ITERATE_BREAK;
}

/**
 * unlink 'waitq' from 'wqset'
 *
 * Conditions:
 *      'waitq' is locked
 *      'wqset' is _not_ locked
 *      may (rarely) spin in prepost clear and drop/re-acquire 'waitq' lock
 *      (see waitq_clear_prepost_locked)
 */
static kern_return_t waitq_unlink_locked(struct waitq *waitq,
                                         struct waitq_set *wqset,
                                         spl_t *s)
{
        uint64_t setid;
        kern_return_t kr;

        setid = wqset->wqset_id;

        if (waitq->waitq_set_id == 0) {
                /*
                 * TODO:
                 * it doesn't belong to anyone, and it has a prepost object?
                 * This is an artifact of not cleaning up after kqueues when
                 * they prepost into select sets...
                 */
                if (waitq->waitq_prepost_id != 0)
                        (void)waitq_clear_prepost_locked(waitq, s);
                return KERN_NOT_IN_SET;
        }

        if (waitq->waitq_set_id == setid) {
                waitq->waitq_set_id = 0;
                /*
                 * This was the only set to which the waitq belonged: we can
                 * safely release the waitq's prepost object. It doesn't
                 * matter if this function drops and re-acquires the lock
                 * because we're not manipulating waitq state any more.
                 */
                (void)waitq_clear_prepost_locked(waitq, s);
                return KERN_SUCCESS;
        }

        /*
         * The waitq was a member of more that 1 set, so we need to
         * handle potentially compressing the link table, and
         * adjusting the waitq->waitq_set_id value.
         *
         * Note: we can't free the waitq's associated prepost object (if any)
         *       because it may be in use by the one or more _other_ sets to
         *       which this queue belongs.
         *
         * Note: This function only handles a single level of the queue linkage.
         *       Removing a waitq from a set to which it does not directly
         *       belong is undefined. For example, if a waitq belonged to set
         *       A, and set A belonged to set B. You can't remove the waitq
         *       from set B.
         */
        kr = walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, waitq->waitq_set_id,
                              SLT_LINK, (void *)&setid, waitq_unlink_cb);

        if (kr == WQ_ITERATE_UNLINKED) {
                struct wq_unlink_ctx ulctx;
                int do_spl = 0;

                kr = KERN_SUCCESS; /* found it and dis-associated it */

                if (!waitq_irq_safe(waitq) && waitq_irq_safe(&wqset->wqset_q)) {
                        *s = splsched();
                        do_spl = 1;
                }
                waitq_set_lock(wqset);
                /*
                 * clear out any prepost from waitq into wqset
                 * TODO: this could be more efficient than a linear search of
                 *       the waitq set's prepost list.
                 */
                ulctx.unlink_wq = waitq;
                ulctx.unlink_wqset = wqset;
                (void)wq_prepost_iterate(wqset->wqset_prepost_id, (void *)&ulctx,
                                         waitq_unlink_prepost_cb);
                waitq_set_unlock(wqset);
                if (do_spl)
                        splx(*s);
        } else {
                kr = KERN_NOT_IN_SET; /* waitq is _not_ associated with wqset */
        }

        return kr;
}

/**
 * unlink 'waitq' from 'wqset'
 *
 * Conditions:
 *      neither 'waitq' nor 'wqset' is locked
 *      may disable and re-enable interrupts
 *      may (rarely) spin in prepost clear
 *      (see waitq_clear_prepost_locked)
 */
kern_return_t waitq_unlink(struct waitq *waitq, struct waitq_set *wqset)
{
        kern_return_t kr = KERN_SUCCESS;
        spl_t s;

        assert(waitqs_is_set(wqset));

        /*
         * we allow the waitq to be invalid because the caller may be trying
         * to clear out old/dirty state
         */
        if (!waitq_valid(waitq))
                return KERN_INVALID_ARGUMENT;

        wqdbg_v("unlink waitq %p from set 0x%llx",
                (void *)VM_KERNEL_UNSLIDE_OR_PERM(waitq), wqset->wqset_id);

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        kr = waitq_unlink_locked(waitq, wqset, &s);

        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

        return kr;
}

/**
 * unlink a waitq from a waitq set, but reference the waitq by its prepost ID
 *
 * Conditions:
 *      'wqset' is unlocked
 *      wqp_id may be valid or invalid
 */
void waitq_unlink_by_prepost_id(uint64_t wqp_id, struct waitq_set *wqset)
{
        struct wq_prepost *wqp;

        disable_preemption();
        wqp = wq_prepost_get(wqp_id);
        if (wqp) {
                struct waitq *wq;
                spl_t s;

                wq = wqp->wqp_wq.wqp_wq_ptr;

                /*
                 * lock the waitq, then release our prepost ID reference, then
                 * unlink the waitq from the wqset: this ensures that we don't
                 * hold a prepost ID reference during the unlink, but we also
                 * complete the unlink operation atomically to avoid a race
                 * with waitq_unlink[_all].
                 */
                if (waitq_irq_safe(wq))
                        s = splsched();
                waitq_lock(wq);
                wq_prepost_put(wqp);

                if (!waitq_valid(wq)) {
                        /* someone already tore down this waitq! */
                        waitq_unlock(wq);
                        if (waitq_irq_safe(wq))
                                splx(s);
                        enable_preemption();
                        return;
                }

                /* this _may_ drop the wq lock, but that's OK */
                waitq_unlink_locked(wq, wqset, &s);

                waitq_unlock(wq);
                if (waitq_irq_safe(wq))
                        splx(s);
        }
        enable_preemption();
        return;
}


/**
 * unlink 'waitq' from all sets to which it belongs
 *
 * Conditions:
 *      'waitq' is locked
 *
 * Notes:
 *      may drop and re-acquire the waitq lock
 *      may (rarely) spin (see waitq_clear_prepost_locked)
 */
kern_return_t waitq_unlink_all_locked(struct waitq *waitq, uint64_t *old_set_id,
                                      spl_t *s, int *dropped_lock)
{
        wqdbg_v("unlink waitq %p from all sets",
                (void *)VM_KERNEL_UNSLIDE_OR_PERM(waitq));

        *old_set_id = 0;

        /* it's not a member of any sets */
        if (waitq->waitq_set_id == 0)
                return KERN_SUCCESS;

        *old_set_id = waitq->waitq_set_id;
        waitq->waitq_set_id = 0;

        /*
         * invalidate the prepost entry for this waitq.
         * This may drop and re-acquire the waitq lock, but that's OK because
         * if it was added to another set and preposted to that set in the
         * time we drop the lock, the state will remain consistent.
         */
        int dropped = waitq_clear_prepost_locked(waitq, s);
        if (dropped_lock)
                *dropped_lock = dropped;

        return KERN_SUCCESS;
}

/**
 * unlink 'waitq' from all sets to which it belongs
 *
 * Conditions:
 *      'waitq' is not locked
 *      may disable and re-enable interrupts
 *      may (rarely) spin
 *      (see waitq_unlink_all_locked, waitq_clear_prepost_locked)
 */
kern_return_t waitq_unlink_all(struct waitq *waitq)
{
        kern_return_t kr = KERN_SUCCESS;
        uint64_t setid = 0;
        spl_t s;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);
        if (waitq_valid(waitq))
                kr = waitq_unlink_all_locked(waitq, &setid, &s, NULL);
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

        if (setid) {
                /*
                 * Walk the link table and invalidate each LINK object that
                 * used to connect this waitq to one or more sets: this works
                 * because SLT_LINK objects are private to each wait queue
                 */
                (void)walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, setid,
                                       SLT_LINK, NULL, waitq_unlink_all_cb);
        }

        return kr;
}


/**
 * unlink all waitqs from 'wqset'
 *
 * Conditions:
 *      'wqset' is not locked
 *      may (rarely) spin/block (see waitq_clear_prepost_locked)
 */
kern_return_t waitq_set_unlink_all(struct waitq_set *wqset)
{
        struct setid_link *link;
        uint64_t prepost_id, set_links_id = 0;
        spl_t spl;

        assert(waitqs_is_set(wqset));

        wqdbg_v("unlink all queues from set 0x%llx", wqset->wqset_id);

        /*
         * This operation does not require interaction with any of the set's
         * constituent wait queues. All we have to do is invalidate the SetID
         */
        if (waitq_irq_safe(&wqset->wqset_q))
                spl = splsched();
        waitq_set_lock(wqset);

        /* invalidate and re-alloc the link object first */
        link = lt_get_link(wqset->wqset_id);

        /* we may have raced with a waitq_set_deinit: handle this */
        if (!link) {
                waitq_set_unlock(wqset);
                return KERN_SUCCESS;
        }

        lt_invalidate(link);

        /* re-alloc the object to get a new generation ID */
        lt_realloc_link(link, SLT_WQS);
        link->sl_wqs.sl_set = wqset;

        wqset->wqset_id = link->sl_set_id.id;
        sl_set_valid(link);
        lt_put_link(link);

        /* clear any preposts attached to this set */
        prepost_id = wqset->wqset_prepost_id;
        wqset->wqset_prepost_id = 0;

        /*
         * clear set linkage and prepost object associated with this set:
         * waitq sets may prepost to other sets if, for example, they are
         * associated with a kqueue which is in a select set.
         *
         * This may drop and re-acquire the set lock, but that's OK because
         * the resulting state will remain consistent.
         */
        waitq_unlink_all_locked(&wqset->wqset_q, &set_links_id, &spl, NULL);

        waitq_set_unlock(wqset);
        if (waitq_irq_safe(&wqset->wqset_q))
                splx(spl);

        /*
         * release all the set link objects
         * (links to other sets to which this set was previously added)
         */
        if (set_links_id)
                (void)walk_setid_links(LINK_WALK_ONE_LEVEL, &wqset->wqset_q,
                                       set_links_id, SLT_LINK, NULL,
                                       waitq_unlink_all_cb);

        /* drop / unlink all the prepost table objects */
        if (prepost_id)
                (void)wq_prepost_iterate(prepost_id, NULL,
                                         wqset_clear_prepost_chain_cb);

        return KERN_SUCCESS;
}


static int waitq_prepost_reserve_cb(struct waitq *waitq, void *ctx,
                                    struct setid_link *link)
{
        uint32_t *num = (uint32_t *)ctx;
        (void)waitq;

        /*
         * In the worst case, we'll have to allocate 2 prepost objects
         * per waitq set (if the set was already preposted by another
         * waitq).
         */
        if (sl_type(link) == SLT_WQS) {
                /*
                 * check to see if the associated waitq actually supports
                 * preposting
                 */
                if (waitq_set_can_prepost(link->sl_wqs.sl_set))
                        *num += 2;
        }
        return WQ_ITERATE_CONTINUE;
}

static int waitq_alloc_prepost_reservation(int nalloc, struct waitq *waitq,
                                           spl_t *s, int *did_unlock,
                                           struct wq_prepost **wqp)
{
        struct wq_prepost *tmp;
        struct wqp_cache *cache;

        *did_unlock = 0;

        /*
         * Before we unlock the waitq, check the per-processor prepost object
         * cache to see if there's enough there for us. If so, do the
         * allocation, keep the lock and save an entire iteration over the set
         * linkage!
         */
        if (waitq) {
                disable_preemption();
                cache = &PROCESSOR_DATA(current_processor(), wqp_cache);
                if (nalloc <= (int)cache->avail)
                        goto do_alloc;
                enable_preemption();

                /* unlock the waitq to perform the allocation */
                *did_unlock = 1;
                waitq_unlock(waitq);
                if (waitq_irq_safe(waitq))
                        splx(*s);
        }

do_alloc:
        tmp = wq_prepost_alloc(WQT_RESERVED, nalloc);
        if (!tmp)
                panic("Couldn't reserve %d preposts for waitq @%p (wqp@%p)",
                      nalloc, waitq, *wqp);
        if (*wqp) {
                /* link the two lists */
                int __assert_only rc;
                rc = wq_prepost_rlink(tmp, *wqp);
                assert(rc == nalloc);
        }
        *wqp = tmp;

        /*
         * If the caller can block, then enforce a minimum-free table element
         * policy here. This helps ensure that we will have enough prepost
         * objects for callers such as selwakeup() that can be called with
         * spin locks held.
         */
        if (get_preemption_level() == 0)
                wq_prepost_ensure_free_space();

        if (waitq) {
                if (*did_unlock == 0) {
                        /* decrement the preemption count if alloc from cache */
                        enable_preemption();
                } else {
                        /* otherwise: re-lock the waitq */
                        if (waitq_irq_safe(waitq))
                                *s = splsched();
                        waitq_lock(waitq);
                }
        }

        return nalloc;
}

static int waitq_count_prepost_reservation(struct waitq *waitq, int extra, int keep_locked)
{
        int npreposts = 0;

        /*
         * If the waitq is not currently part of a set, and we're not asked to
         * keep the waitq locked then we'll want to have 3 in reserve
         * just-in-case it becomes part of a set while we unlock and reserve.
         * We may need up to 1 object for the waitq, and 2 for the set.
         */
        if (waitq->waitq_set_id == 0) {
                npreposts = 3;
        } else {
                /* this queue has never been preposted before */
                if (waitq->waitq_prepost_id == 0)
                        npreposts = 3;

                /*
                 * Walk the set of table linkages associated with this waitq
                 * and count the worst-case number of prepost objects that
                 * may be needed during a wakeup_all. We can walk this without
                 * locking each set along the way because the table-based IDs
                 * disconnect us from the set pointers themselves, and the
                 * table walking is careful to read the setid values only once.
                 * Locking each set up the chain also doesn't guarantee that
                 * their membership won't change between the time we unlock
                 * that set and when we actually go to prepost, so our
                 * situation is no worse than before and we've alleviated lock
                 * contention on any sets to which this waitq belongs.
                 */
                (void)walk_setid_links(LINK_WALK_FULL_DAG_UNLOCKED,
                                       waitq, waitq->waitq_set_id,
                                       SLT_WQS, (void *)&npreposts,
                                       waitq_prepost_reserve_cb);
        }

        if (extra > 0)
                npreposts += extra;

        if (npreposts == 0 && !keep_locked) {
                /*
                 * If we get here, we were asked to reserve some prepost
                 * objects for a waitq that's previously preposted, and is not
                 * currently a member of any sets. We have also been
                 * instructed to unlock the waitq when we're done. In this
                 * case, we pre-allocated enough reserved objects to handle
                 * the case where the waitq gets added to a single set when
                 * the lock is released.
                 */
                npreposts = 3;
        }

        return npreposts;
}


/**
 * pre-allocate prepost objects for 'waitq'
 *
 * Conditions:
 *      'waitq' is not locked
 *
 * Returns:
 *      panic on error
 *
 *      0 on success, '*reserved' is set to the head of a singly-linked
 *      list of pre-allocated prepost objects.
 *
 * Notes:
 *      If 'lock_state' is WAITQ_KEEP_LOCKED, this function performs the pre-allocation
 *      atomically and returns 'waitq' locked. If the waitq requires
 *      interrupts to be disabled, then the output parameter 's' is set to the
 *      previous interrupt state (from splsched), and the caller is
 *      responsible to call splx().
 *
 *      This function attempts to pre-allocate precisely enough prepost
 *      objects based on the current set membership of 'waitq'. If the
 *      operation is performed atomically, then the caller
 *      is guaranteed to have enough pre-allocated prepost object to avoid
 *      any (rare) blocking in the wakeup path.
 */
uint64_t waitq_prepost_reserve(struct waitq *waitq, int extra,
                               waitq_lock_state_t lock_state, spl_t *s)
{
        uint64_t reserved = 0;
        uint64_t prev_setid = 0, prev_prepostid = 0;
        struct wq_prepost *wqp = NULL;
        int nalloc = 0, npreposts = 0;
        int keep_locked = (lock_state == WAITQ_KEEP_LOCKED);
        int unlocked = 0;

        if (s)
                *s = 0;

        wqdbg_v("Attempting to reserve prepost linkages for waitq %p (extra:%d)",
                (void *)VM_KERNEL_UNSLIDE_OR_PERM(waitq), extra);

        if (waitq == NULL && extra > 0) {
                /*
                 * Simple prepost object allocation:
                 * we'll add 2 more because the waitq might need an object,
                 * and the set itself may need a new POST object in addition
                 * to the number of preposts requested by the caller
                 */
                nalloc = waitq_alloc_prepost_reservation(extra + 2, NULL, NULL,
                                                         &unlocked, &wqp);
                assert(nalloc == extra + 2);
                return wqp->wqp_prepostid.id;
        }

        assert(lock_state == WAITQ_KEEP_LOCKED || lock_state == WAITQ_UNLOCK);

        if (waitq_irq_safe(waitq))
                *s = splsched();
        waitq_lock(waitq);

        /* global queues are never part of any sets */
        if (waitq_is_global(waitq)) {
                if (keep_locked)
                        goto out;
                goto out_unlock;
        }

        /* remember the set ID that we started with */
        prev_setid = waitq->waitq_set_id;
        prev_prepostid = waitq->waitq_prepost_id;

        /*
         * If the waitq is not part of a set, and we're asked to
         * keep the set locked, then we don't have to reserve
         * anything!
         */
        if (prev_setid == 0 && keep_locked)
                goto out;

        npreposts = waitq_count_prepost_reservation(waitq, extra, keep_locked);

        /* nothing for us to do! */
        if (npreposts == 0) {
                if (keep_locked)
                        goto out;
                goto out_unlock;
        }

try_alloc:
        /* this _may_ unlock and relock the waitq! */
        nalloc = waitq_alloc_prepost_reservation(npreposts, waitq, s,
                                                 &unlocked, &wqp);

        if (!unlocked) {
                /* allocation held the waitq lock: we'd done! */
                if (keep_locked)
                        goto out;
                goto out_unlock;
        }

        /*
         * Before we return, if the allocation had to unlock the waitq, we
         * must check one more time to see if we have enough. If not, we'll
         * try to allocate the difference. If the caller requests it, we'll
         * also leave the waitq locked so that the use of the pre-allocated
         * prepost objects can be guaranteed to be enough if a wakeup_all is
         * performed before unlocking the waitq.
         */

        /*
         * If the waitq is no longer associated with a set, or if the waitq's
         * set/prepostid has not changed since we first walked its linkage,
         * we're done.
         */
        if ((waitq->waitq_set_id == 0) ||
            (waitq->waitq_set_id == prev_setid &&
             waitq->waitq_prepost_id == prev_prepostid)) {
                if (keep_locked)
                        goto out;
                goto out_unlock;
        }

        npreposts = waitq_count_prepost_reservation(waitq, extra, keep_locked);

        if (npreposts > nalloc) {
                prev_setid = waitq->waitq_set_id;
                prev_prepostid = waitq->waitq_prepost_id;
                npreposts = npreposts - nalloc; /* only allocate the diff */
                goto try_alloc;
        }

        if (keep_locked)
                goto out;

out_unlock:
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(*s);
out:
        if (wqp)
                reserved = wqp->wqp_prepostid.id;

        return reserved;
}

/**
 * release a linked list of prepost objects allocated via _prepost_reserve
 *
 * Conditions:
 *      may (rarely) spin waiting for prepost table growth memcpy
 */
void waitq_prepost_release_reserve(uint64_t id)
{
        struct wq_prepost *wqp;

        wqdbg_v("releasing reserved preposts starting at: 0x%llx", id);

        wqp = wq_prepost_rfirst(id);
        if (!wqp)
                return;

        wq_prepost_release_rlist(wqp);
}


/**
 * clear all preposts from 'wqset'
 *
 * Conditions:
 *      'wqset' is not locked
 */
void waitq_set_clear_preposts(struct waitq_set *wqset)
{
        uint64_t prepost_id;
        spl_t spl;

        assert(waitqs_is_set(wqset));

        wqdbg_v("Clearing all preposted queues on waitq_set: 0x%llx",
                wqset->wqset_id);

        if (waitq_irq_safe(&wqset->wqset_q))
                spl = splsched();
        waitq_set_lock(wqset);
        prepost_id = wqset->wqset_prepost_id;
        wqset->wqset_prepost_id = 0;
        waitq_set_unlock(wqset);
        if (waitq_irq_safe(&wqset->wqset_q))
                splx(spl);

        /* drop / unlink all the prepost table objects */
        if (prepost_id)
                (void)wq_prepost_iterate(prepost_id, NULL,
                                         wqset_clear_prepost_chain_cb);
}


/* ----------------------------------------------------------------------
 *
 * Iteration: waitq -> sets / waitq_set -> preposts
 *
 * ---------------------------------------------------------------------- */

struct wq_it_ctx {
        void *input;
        void *ctx;
        waitq_iterator_t it;

        spl_t *spl;
};

static int waitq_iterate_sets_cb(struct waitq *waitq, void *ctx,
                                 struct setid_link *link)
{
        struct wq_it_ctx *wctx = (struct wq_it_ctx *)(ctx);
        struct waitq_set *wqset;
        int ret;
        spl_t spl;

        (void)waitq;
        assert(sl_type(link) == SLT_WQS);

        /*
         * the waitq is locked, so we can just take the set lock
         * and call the iterator function
         */
        wqset = link->sl_wqs.sl_set;
        assert(wqset != NULL);

        if (!waitq_irq_safe(waitq) && waitq_irq_safe(&wqset->wqset_q))
                spl = splsched();
        waitq_set_lock(wqset);

        ret = wctx->it(wctx->ctx, (struct waitq *)wctx->input, wqset);

        waitq_set_unlock(wqset);
        if (!waitq_irq_safe(waitq) && waitq_irq_safe(&wqset->wqset_q))
                splx(spl);

        return ret;
}

/**
 * call external iterator function for each prepost object in wqset
 *
 * Conditions:
 *      Called from wq_prepost_foreach_locked
 *      (wqset locked, waitq _not_ locked)
 */
static int wqset_iterate_prepost_cb(struct waitq_set *wqset, void *ctx,
                                    struct wq_prepost *wqp, struct waitq *waitq)
{
        struct wq_it_ctx *wctx = (struct wq_it_ctx *)(ctx);
        uint64_t wqp_id;
        int ret;
        spl_t s;

        (void)wqp;

        /*
         * This is a bit tricky. The 'wqset' is locked, but the 'waitq' is not.
         * Taking the 'waitq' lock is a lock order violation, so we need to be
         * careful. We also must realize that we may have taken a reference to
         * the 'wqp' just as the associated waitq was being torn down (or
         * clearing all its preposts) - see waitq_clear_prepost_locked(). If
         * the 'wqp' is valid and we can get the waitq lock, then we are good
         * to go. If not, we need to back off, check that the 'wqp' hasn't
         * been invalidated, and try to re-take the locks.
         */
        if (waitq_irq_safe(waitq))
                s = splsched();
        if (waitq_lock_try(waitq))
                goto call_iterator;

        if (waitq_irq_safe(waitq))
                splx(s);

        if (!wqp_is_valid(wqp))
                return WQ_ITERATE_RESTART;

        /* We are passed a prepost object with a reference on it. If neither
         * the waitq set nor the waitq require interrupts disabled, then we
         * may block on the delay(1) call below. We can't hold a prepost
         * object reference while blocking, so we have to give that up as well
         * and re-acquire it when we come back.
         */
        wqp_id = wqp->wqp_prepostid.id;
        wq_prepost_put(wqp);
        waitq_set_unlock(wqset);
        wqdbg_v("dropped set:%p lock waiting for wqp:%p (0x%llx -> wq:%p)",
                wqset, wqp, wqp->wqp_prepostid.id, waitq);
        delay(1);
        waitq_set_lock(wqset);
        wqp = wq_prepost_get(wqp_id);
        if (!wqp)
                /* someone cleared preposts while we slept! */
                return WQ_ITERATE_DROPPED;

        /*
         * TODO:
         * This differs slightly from the logic in ipc_mqueue.c:
         * ipc_mqueue_receive_on_thread(). There, if the waitq lock
         * can't be obtained, the prepost link is placed on the back of
         * the chain, and the iteration starts from the beginning. Here,
         * we just restart from the beginning.
         */
        return WQ_ITERATE_RESTART;

call_iterator:
        if (!wqp_is_valid(wqp)) {
                ret = WQ_ITERATE_RESTART;
                goto out_unlock;
        }

        /* call the external callback */
        ret = wctx->it(wctx->ctx, waitq, wqset);

        if (ret == WQ_ITERATE_BREAK_KEEP_LOCKED) {
                ret = WQ_ITERATE_BREAK;
                if (wctx->spl)
                        *(wctx->spl) = s;
                goto out;
        }

out_unlock:
        waitq_unlock(waitq);
        if (waitq_irq_safe(waitq))
                splx(s);

out:
        return ret;
}

/**
 * iterator over all sets to which the given waitq has been linked
 *
 * Conditions:
 *      'waitq' is locked
 */
int waitq_iterate_sets(struct waitq *waitq, void *ctx, waitq_iterator_t it)
{
        int ret;
        struct wq_it_ctx wctx = {
                .input = (void *)waitq,
                .ctx = ctx,
                .it = it,
        };
        if (!it || !waitq)
                return KERN_INVALID_ARGUMENT;

        ret = walk_setid_links(LINK_WALK_ONE_LEVEL, waitq, waitq->waitq_set_id,
                               SLT_WQS, (void *)&wctx, waitq_iterate_sets_cb);
        if (ret == WQ_ITERATE_CONTINUE)
                ret = WQ_ITERATE_SUCCESS;
        return ret;
}

/**
 * iterator over all preposts in the given wqset
 *
 * Conditions:
 *      'wqset' is locked
 */
int waitq_set_iterate_preposts(struct waitq_set *wqset,
                               void *ctx, waitq_iterator_t it, spl_t *s)
{
        struct wq_it_ctx wctx = {
                .input = (void *)wqset,
                .ctx = ctx,
                .it = it,
                .spl = s,
        };
        if (!it || !wqset)
                return WQ_ITERATE_INVALID;

        assert(waitq_held(&wqset->wqset_q));

        return wq_prepost_foreach_locked(wqset, (void *)&wctx,
                                         wqset_iterate_prepost_cb);
}


/* ----------------------------------------------------------------------
 *
 * Higher-level APIs
 *
 * ---------------------------------------------------------------------- */

/**
 * declare a thread's intent to wait on 'waitq' for 'wait_event'
 *
 * Conditions:
 *      'waitq' is not locked
 *      will disable and re-enable interrupts while locking current_thread()
 */
wait_result_t waitq_assert_wait64(struct waitq *waitq,
                                  event64_t wait_event,
                                  wait_interrupt_t interruptible,
                                  uint64_t deadline)
{
        wait_result_t ret;
        thread_t thread = current_thread();
        spl_t s;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        if (!waitq_irq_safe(waitq))
                s = splsched();
        thread_lock(thread);

        ret = waitq_assert_wait64_locked(waitq, wait_event, interruptible,
                                         TIMEOUT_URGENCY_SYS_NORMAL,
                                         deadline, TIMEOUT_NO_LEEWAY, thread);

        thread_unlock(thread);
        waitq_unlock(waitq);

        splx(s);

        return ret;
}

/**
 * declare a thread's intent to wait on 'waitq' for 'wait_event'
 *
 * Conditions:
 *      'waitq' is not locked
 *      will disable and re-enable interrupts while locking current_thread()
 */
wait_result_t waitq_assert_wait64_leeway(struct waitq *waitq,
                                         event64_t wait_event,
                                         wait_interrupt_t interruptible,
                                         wait_timeout_urgency_t urgency,
                                         uint64_t deadline,
                                         uint64_t leeway)
{
        wait_result_t ret;
        thread_t thread = current_thread();
        spl_t s;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        if (!waitq_irq_safe(waitq))
                s = splsched();
        thread_lock(thread);

        ret = waitq_assert_wait64_locked(waitq, wait_event, interruptible,
                                         urgency, deadline, leeway, thread);

        thread_unlock(thread);
        waitq_unlock(waitq);

        splx(s);

        return ret;
}

/**
 * wakeup a single thread from a waitq that's waiting for a given event
 *
 * Conditions:
 *      'waitq' is not locked
 *      may (rarely) block if 'waitq' is non-global and a member of 1 or more sets
 *      may disable and re-enable interrupts
 *
 * Notes:
 *      will _not_ block if waitq is global (or not a member of any set)
 */
kern_return_t waitq_wakeup64_one(struct waitq *waitq, event64_t wake_event,
                                 wait_result_t result, int priority)
{
        kern_return_t kr;
        uint64_t reserved_preposts = 0;
        spl_t spl;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        /* NOTE: this will _not_ reserve anything if waitq is global */
        reserved_preposts = waitq_prepost_reserve(waitq, 0,
                                                  WAITQ_KEEP_LOCKED, &spl);

        /* waitq is locked upon return */
        kr = waitq_wakeup64_one_locked(waitq, wake_event, result,
                                       &reserved_preposts, priority, WAITQ_UNLOCK);

        if (waitq_irq_safe(waitq))
                splx(spl);

        /* release any left-over prepost object (won't block/lock anything) */
        waitq_prepost_release_reserve(reserved_preposts);

        return kr;
}

/**
 * wakeup all threads from a waitq that are waiting for a given event
 *
 * Conditions:
 *      'waitq' is not locked
 *      may (rarely) block if 'waitq' is non-global and a member of 1 or more sets
 *      may disable and re-enable interrupts
 *
 * Notes:
 *      will _not_ block if waitq is global (or not a member of any set)
 */
kern_return_t waitq_wakeup64_all(struct waitq *waitq,
                                 event64_t wake_event,
                                 wait_result_t result,
                                 int priority)
{
        kern_return_t ret;
        uint64_t reserved_preposts = 0;
        spl_t s;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        /* keep waitq locked upon return */
        /* NOTE: this will _not_ reserve anything if waitq is global */
        reserved_preposts = waitq_prepost_reserve(waitq, 0,
                                                  WAITQ_KEEP_LOCKED, &s);

        /* waitq is locked */

        ret = waitq_wakeup64_all_locked(waitq, wake_event, result,
                                        &reserved_preposts, priority,
                                        WAITQ_UNLOCK);

        if (waitq_irq_safe(waitq))
                splx(s);

        waitq_prepost_release_reserve(reserved_preposts);

        return ret;

}

/**
 * wakeup a specific thread iff it's waiting on 'waitq' for 'wake_event'
 *
 * Conditions:
 *      'waitq' is not locked
 *
 * Notes:
 *      May temporarily disable and re-enable interrupts
 */
kern_return_t waitq_wakeup64_thread(struct waitq *waitq,
                                    event64_t wake_event,
                                    thread_t thread,
                                    wait_result_t result)
{
        kern_return_t ret;
        spl_t s, th_spl;

        if (!waitq_valid(waitq))
                panic("Invalid waitq: %p", waitq);

        if (waitq_irq_safe(waitq))
                s = splsched();
        waitq_lock(waitq);

        ret = waitq_select_thread_locked(waitq, wake_event, thread, &th_spl);
        /* on success, returns 'thread' locked */

        waitq_unlock(waitq);

        if (ret == KERN_SUCCESS) {
                ret = thread_go(thread, result);
                assert(ret == KERN_SUCCESS);
                thread_unlock(thread);
                splx(th_spl);
                waitq_stats_count_wakeup(waitq);
        } else {
                ret = KERN_NOT_WAITING;
                waitq_stats_count_fail(waitq);
        }

        if (waitq_irq_safe(waitq))
                splx(s);

        return ret;
}