xnu-3789.21.4.tar.gz

[apple/xnu.git] / bsd / kern / sys_ulock.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@
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/ioctl.h>
#include <sys/file_internal.h>
#include <sys/proc_internal.h>
#include <sys/kernel.h>
#include <sys/guarded.h>
#include <sys/stat.h>
#include <sys/malloc.h>
#include <sys/sysproto.h>
#include <sys/pthread_shims.h>

#include <mach/mach_types.h>

#include <kern/cpu_data.h>
#include <kern/mach_param.h>
#include <kern/kern_types.h>
#include <kern/assert.h>
#include <kern/kalloc.h>
#include <kern/thread.h>
#include <kern/clock.h>
#include <kern/ledger.h>
#include <kern/policy_internal.h>
#include <kern/task.h>
#include <kern/telemetry.h>
#include <kern/waitq.h>
#include <kern/sched_prim.h>
#include <kern/zalloc.h>

#include <pexpert/pexpert.h>

#define XNU_TEST_BITMAP
#include <kern/bits.h>

#include <sys/ulock.h>

/*
 * How ulock promotion works:
 *
 * There’s a requested policy field on every thread called ‘promotions’, which
 * expresses which ulock promotions are happening to this thread.
 * The promotion priority saturates until the promotion count goes to 0.
 *
 * We also track effective promotion qos, which is the qos before clamping.
 * This value is used for promoting a thread that another thread is waiting on,
 * so that the lock owner reinflates to the right priority after unclamping.
 *
 * This also works for non-QoS threads, which can donate base priority to QoS
 * and non-QoS threads alike.
 *
 * ulock wait applies a promotion to the owner communicated through
 * UL_UNFAIR_LOCK as waiters block, and that promotion is saturated as long as
 * there is still an owner.  In ulock wake, if the waker is still the owner,
 * then it clears its ownership and drops the boost.  It does NOT transfer
 * ownership/priority boost to the new thread.  Instead, it selects the
 * waiting thread with the highest base priority to be woken next, and
 * relies on that thread to carry the torch for the other waiting threads.
 */

static lck_grp_t *ull_lck_grp;
static lck_mtx_t ull_table_lock;

#define ull_global_lock()       lck_mtx_lock(&ull_table_lock)
#define ull_global_unlock()     lck_mtx_unlock(&ull_table_lock)

#define ull_lock(ull)           lck_mtx_lock(&ull->ull_lock)
#define ull_unlock(ull)         lck_mtx_unlock(&ull->ull_lock)
#define ull_assert_owned(ull)   LCK_MTX_ASSERT(&ull->ull_lock, LCK_MTX_ASSERT_OWNED)

typedef struct __attribute__((packed)) {
        user_addr_t     ulk_addr;
        pid_t           ulk_pid;
} ulk_t;

inline static bool
ull_key_match(ulk_t *a, ulk_t *b)
{
        return ((a->ulk_pid == b->ulk_pid) &&
                (a->ulk_addr == b->ulk_addr));
}

typedef struct ull {
        /*
         * ull_owner is the most recent known value for the owner of this ulock
         * i.e. it may be out of date WRT the real value in userspace.
         */
        thread_t        ull_owner; /* holds +1 thread reference */
        ulk_t           ull_key;
        ulk_t           ull_saved_key;
        lck_mtx_t       ull_lock;
        int32_t         ull_nwaiters;
        int32_t         ull_max_nwaiters;
        int32_t         ull_refcount;
        struct promote_token ull_promote_token;
        queue_chain_t   ull_hash_link;
        uint8_t         ull_opcode;
} ull_t;

static const bool ull_debug = false;

extern void ulock_initialize(void);

#define ULL_MUST_EXIST  0x0001
static ull_t *ull_get(ulk_t *, uint32_t);
static void ull_put(ull_t *);

static thread_t ull_promote_owner_locked(ull_t* ull, thread_t thread);

#if DEVELOPMENT || DEBUG
static int ull_simulate_copyin_fault = 0;
static int ull_panic_on_corruption = 0;

static void
ull_dump(ull_t *ull)
{
        kprintf("ull\t%p\n", ull);
        kprintf("ull_key.ulk_pid\t%d\n", ull->ull_key.ulk_pid);
        kprintf("ull_key.ulk_addr\t%p\n", (void *)(ull->ull_key.ulk_addr));
        kprintf("ull_saved_key.ulk_pid\t%d\n", ull->ull_saved_key.ulk_pid);
        kprintf("ull_saved_key.ulk_addr\t%p\n", (void *)(ull->ull_saved_key.ulk_addr));
        kprintf("ull_nwaiters\t%d\n", ull->ull_nwaiters);
        kprintf("ull_max_nwaiters\t%d\n", ull->ull_max_nwaiters);
        kprintf("ull_refcount\t%d\n", ull->ull_refcount);
        kprintf("ull_opcode\t%d\n\n", ull->ull_opcode);
        kprintf("ull_owner\t0x%llx\n\n", thread_tid(ull->ull_owner));
        kprintf("ull_promote_token\t%d, %d\n\n", ull->ull_promote_token.pt_basepri, ull->ull_promote_token.pt_qos);
}
#endif

static int ull_hash_buckets;
static queue_head_t *ull_bucket;
static uint32_t ull_nzalloc = 0;
static zone_t ull_zone;

static __inline__ uint32_t
ull_hash_index(char *key, size_t length)
{
        uint32_t hash = jenkins_hash(key, length);

        hash &= (ull_hash_buckets - 1);

        return hash;
}

/* Ensure that the key structure is packed,
 * so that no undefined memory is passed to
 * ull_hash_index()
 */
static_assert(sizeof(ulk_t) == sizeof(user_addr_t) + sizeof(pid_t));

#define ULL_INDEX(keyp) ull_hash_index((char *)keyp, sizeof *keyp)

void
ulock_initialize(void)
{
        ull_lck_grp = lck_grp_alloc_init("ulocks", NULL);
        lck_mtx_init(&ull_table_lock, ull_lck_grp, NULL);

        assert(thread_max > 16);
        /* Size ull_hash_buckets based on thread_max.
         * Round up to nearest power of 2, then divide by 4
         */
        ull_hash_buckets = (1 << (bit_ceiling(thread_max) - 2));

        kprintf("%s>thread_max=%d, ull_hash_buckets=%d\n", __FUNCTION__, thread_max, ull_hash_buckets);
        assert(ull_hash_buckets >= thread_max/4);

        ull_bucket = (queue_head_t *)kalloc(sizeof(queue_head_t) * ull_hash_buckets);
        assert(ull_bucket != NULL);

        for (int i = 0; i < ull_hash_buckets; i++) {
                queue_init(&ull_bucket[i]);
        }

        ull_zone = zinit(sizeof(ull_t),
                         thread_max * sizeof(ull_t),
                         0, "ulocks");

        zone_change(ull_zone, Z_NOENCRYPT, TRUE);

#if DEVELOPMENT || DEBUG
        if (!PE_parse_boot_argn("ulock_panic_on_corruption",
                        &ull_panic_on_corruption, sizeof(ull_panic_on_corruption))) {
                ull_panic_on_corruption = 0;
        }
#endif
}

#if DEVELOPMENT || DEBUG
/* Count the number of hash entries for a given pid.
 * if pid==0, dump the whole table.
 */
static int
ull_hash_dump(pid_t pid)
{
        int count = 0;
        ull_global_lock();
        if (pid == 0) {
                kprintf("%s>total number of ull_t allocated %d\n", __FUNCTION__, ull_nzalloc);
                kprintf("%s>BEGIN\n", __FUNCTION__);
        }
        for (int i = 0; i < ull_hash_buckets; i++) {
                if (!queue_empty(&ull_bucket[i])) {
                        ull_t *elem;
                        if (pid == 0) {
                                kprintf("%s>index %d:\n", __FUNCTION__, i);
                        }
                        qe_foreach_element(elem, &ull_bucket[i], ull_hash_link) {
                                if ((pid == 0) || (pid == elem->ull_key.ulk_pid)) {
                                        ull_dump(elem);
                                        count++;
                                }
                        }
                }
        }
        if (pid == 0) {
                kprintf("%s>END\n", __FUNCTION__);
                ull_nzalloc = 0;
        }
        ull_global_unlock();
        return count;
}
#endif

static ull_t *
ull_alloc(ulk_t *key)
{
        ull_t *ull = (ull_t *)zalloc(ull_zone);
        assert(ull != NULL);

        ull->ull_refcount = 1;
        ull->ull_key = *key;
        ull->ull_saved_key = *key;
        ull->ull_nwaiters = 0;
        ull->ull_max_nwaiters = 0;
        ull->ull_opcode = 0;

        ull->ull_owner = THREAD_NULL;
        ull->ull_promote_token = PROMOTE_TOKEN_INIT;

        lck_mtx_init(&ull->ull_lock, ull_lck_grp, NULL);

        ull_nzalloc++;
        return ull;
}

static void
ull_free(ull_t *ull)
{
        assert(ull->ull_owner == THREAD_NULL);

        lck_mtx_assert(&ull->ull_lock, LCK_ASSERT_NOTOWNED);

        lck_mtx_destroy(&ull->ull_lock, ull_lck_grp);

        zfree(ull_zone, ull);
}

/* Finds an existing ulock structure (ull_t), or creates a new one.
 * If MUST_EXIST flag is set, returns NULL instead of creating a new one.
 * The ulock structure is returned with ull_lock locked
 *
 * TODO: Per-bucket lock to reduce contention on global lock
 */
static ull_t *
ull_get(ulk_t *key, uint32_t flags)
{
        ull_t *ull = NULL;
        uint i = ULL_INDEX(key);
        ull_t *elem;
        ull_global_lock();
        qe_foreach_element(elem, &ull_bucket[i], ull_hash_link) {
                ull_lock(elem);
                if (ull_key_match(&elem->ull_key, key)) {
                        ull = elem;
                        break;
                } else {
                        ull_unlock(elem);
                }
        }
        if (ull == NULL) {
                if (flags & ULL_MUST_EXIST) {
                        /* Must already exist (called from wake) */
                        ull_global_unlock();
                        return NULL;
                }

                /* NRG maybe drop the ull_global_lock before the kalloc,
                 * then take the lock and check again for a key match
                 * and either use the new ull_t or free it.
                 */

                ull = ull_alloc(key);

                if (ull == NULL) {
                        ull_global_unlock();
                        return NULL;
                }

                ull_lock(ull);

                enqueue(&ull_bucket[i], &ull->ull_hash_link);
        }

        ull->ull_refcount++;

        ull_global_unlock();

        return ull; /* still locked */
}

/*
 * Must be called with ull_lock held
 */
static void
ull_put(ull_t *ull)
{
        ull_assert_owned(ull);
        int refcount = --ull->ull_refcount;
        assert(refcount == 0 ? (ull->ull_key.ulk_pid == 0 && ull->ull_key.ulk_addr == 0) : 1);
        ull_unlock(ull);

        if (refcount > 0) {
                return;
        }

        ull_global_lock();
        remqueue(&ull->ull_hash_link);
        ull_global_unlock();

#if DEVELOPMENT || DEBUG
        if (ull_debug) {
                kprintf("%s>", __FUNCTION__);
                ull_dump(ull);
        }
#endif
        ull_free(ull);
}

int
ulock_wait(struct proc *p, struct ulock_wait_args *args, int32_t *retval)
{
        uint opcode = args->operation & UL_OPCODE_MASK;
        uint flags = args->operation & UL_FLAGS_MASK;
        int ret = 0;
        thread_t self = current_thread();
        int id = thread_tid(self);
        ulk_t key;

        /* involved threads - each variable holds +1 ref if not null */
        thread_t owner_thread   = THREAD_NULL;
        thread_t old_owner      = THREAD_NULL;
        thread_t old_lingering_owner = THREAD_NULL;
        sched_call_t workq_callback = NULL;

        if (ull_debug) {
                kprintf("[%d]%s>ENTER opcode %d addr %llx value %llx timeout %d flags %x\n", id, __FUNCTION__, opcode, (unsigned long long)(args->addr), args->value, args->timeout, flags);
        }

        if ((flags & ULF_WAIT_MASK) != flags) {
                ret = EINVAL;
                goto munge_retval;
        }

        boolean_t set_owner = FALSE;

        switch (opcode) {
        case UL_UNFAIR_LOCK:
                set_owner = TRUE;
                break;
        case UL_COMPARE_AND_WAIT:
                break;
        default:
                if (ull_debug) {
                        kprintf("[%d]%s>EINVAL opcode %d addr 0x%llx flags 0x%x\n",
                                id, __FUNCTION__, opcode,
                                (unsigned long long)(args->addr), flags);
                }
                ret = EINVAL;
                goto munge_retval;
        }

        /* 32-bit lock type for UL_COMPARE_AND_WAIT and UL_UNFAIR_LOCK */
        uint32_t value = 0;

        if ((args->addr == 0) || (args->addr % _Alignof(_Atomic(typeof(value))))) {
                ret = EINVAL;
                goto munge_retval;
        }

        key.ulk_pid = p->p_pid;
        key.ulk_addr = args->addr;

        if (flags & ULF_WAIT_WORKQ_DATA_CONTENTION) {
                workq_callback = workqueue_get_sched_callback();
                workq_callback = thread_disable_sched_call(self, workq_callback);
        }

        ull_t *ull = ull_get(&key, 0);
        if (ull == NULL) {
                ret = ENOMEM;
                goto munge_retval;
        }
        /* ull is locked */

        ull->ull_nwaiters++;

        if (ull->ull_nwaiters > ull->ull_max_nwaiters) {
                ull->ull_max_nwaiters = ull->ull_nwaiters;
        }

        if (ull->ull_opcode == 0) {
                ull->ull_opcode = opcode;
        } else if (ull->ull_opcode != opcode) {
                ull_unlock(ull);
                ret = EDOM;
                goto out;
        }

        /*
         * We don't want this copyin to get wedged behind VM operations,
         * but we have to read the userspace value under the ull lock for correctness.
         *
         * Until <rdar://problem/24999882> exists,
         * fake it by disabling preemption across copyin, which forces any
         * vm_fault we encounter to fail.
         */
        uint64_t val64; /* copyin_word always zero-extends to 64-bits */

        disable_preemption();
        int copy_ret = copyin_word(args->addr, &val64, sizeof(value));
        enable_preemption();

        value = (uint32_t)val64;

#if DEVELOPMENT || DEBUG
        /* Occasionally simulate copyin finding the user address paged out */
        if (((ull_simulate_copyin_fault == p->p_pid) || (ull_simulate_copyin_fault == 1)) && (copy_ret == 0)) {
                static _Atomic int fault_inject = 0;
                if (__c11_atomic_fetch_add(&fault_inject, 1, __ATOMIC_RELAXED) % 73 == 0) {
                        copy_ret = EFAULT;
                }
        }
#endif
        if (copy_ret != 0) {
                ull_unlock(ull);

                /* copyin() will return an error if the access to the user addr would have faulted,
                 * so just return and let the user level code fault it in.
                 */
                ret = copy_ret;
                goto out;
        }

        if (value != args->value) {
                /* Lock value has changed from expected so bail out */
                ull_unlock(ull);
                if (ull_debug) {
                        kprintf("[%d]%s>Lock value %d has changed from expected %d so bail out\n",
                                id, __FUNCTION__, value, (uint32_t)(args->value));
                }
                goto out;
        }

        if (set_owner) {
                mach_port_name_t owner_name = ulock_owner_value_to_port_name(args->value);
                owner_thread = port_name_to_thread_for_ulock(owner_name);

                /* HACK: don't bail on MACH_PORT_DEAD, to avoid blowing up the no-tsd pthread lock */
                if (owner_name != MACH_PORT_DEAD && owner_thread == THREAD_NULL) {
#if DEBUG || DEVELOPMENT
                        if (ull_panic_on_corruption) {
                                if (flags & ULF_NO_ERRNO) {
                                        // ULF_NO_ERRNO is used by libplatform ulocks, but not libdispatch ones.
                                        // Don't panic on libdispatch ulock corruptions; the userspace likely
                                        // mismanaged a dispatch queue.
                                        panic("ulock_wait: ulock is corrupted; value=0x%x, ull=%p",
                                                        (uint32_t)(args->value), ull);
                                }
                        }
#endif
                        /*
                         * Translation failed - even though the lock value is up to date,
                         * whatever was stored in the lock wasn't actually a thread port.
                         */
                        ull_unlock(ull);
                        ret = EOWNERDEAD;
                        goto out;
                }
                /* owner_thread has a +1 reference */

                /*
                 * At this point, I know:
                 * a) owner_thread is definitely the current owner, because I just read the value
                 * b) owner_thread is either:
                 *      i) holding the user lock or
                 *      ii) has just unlocked the user lock after I looked
                 *              and is heading toward the kernel to call ull_wake.
                 *              If so, it's going to have to wait for the ull mutex.
                 *
                 * Therefore, I can promote its priority to match mine, and I can rely on it to
                 * come by later to issue the wakeup and lose its promotion.
                 */

                old_owner = ull_promote_owner_locked(ull, owner_thread);
        }

        wait_result_t wr;
        uint32_t timeout = args->timeout;
        if (timeout) {
                wr = assert_wait_timeout((event_t)ull, THREAD_ABORTSAFE, timeout, NSEC_PER_USEC);
        } else {
                wr = assert_wait((event_t)ull, THREAD_ABORTSAFE);
        }

        ull_unlock(ull);

        if (ull_debug) {
                kprintf("[%d]%s>after assert_wait() returned %d\n", id, __FUNCTION__, wr);
        }

        if (set_owner && owner_thread != THREAD_NULL && wr == THREAD_WAITING) {
                wr = thread_handoff(owner_thread);
                /* owner_thread ref is consumed */
                owner_thread = THREAD_NULL;
        } else {
                /* NRG At some point this should be a continuation based block, so that we can avoid saving the full kernel context. */
                wr = thread_block(NULL);
        }
        if (ull_debug) {
                kprintf("[%d]%s>thread_block() returned %d\n", id, __FUNCTION__, wr);
        }
        switch (wr) {
        case THREAD_AWAKENED:
                break;
        case THREAD_TIMED_OUT:
                ret = ETIMEDOUT;
                break;
        case THREAD_INTERRUPTED:
        case THREAD_RESTART:
        default:
                ret = EINTR;
                break;
        }

out:
        ull_lock(ull);
        *retval = --ull->ull_nwaiters;
        if (ull->ull_nwaiters == 0) {
                /*
                 * If the wait was canceled early, we might need to
                 * clear out the lingering owner reference before
                 * freeing the ull.
                 */
                if (ull->ull_owner != THREAD_NULL) {
                        old_lingering_owner = ull_promote_owner_locked(ull, THREAD_NULL);
                }

                assert(ull->ull_owner == THREAD_NULL);

                ull->ull_key.ulk_pid = 0;
                ull->ull_key.ulk_addr = 0;
                ull->ull_refcount--;
                assert(ull->ull_refcount > 0);
        }
        ull_put(ull);

        if (owner_thread != THREAD_NULL) {
                thread_deallocate(owner_thread);
        }

        if (old_owner != THREAD_NULL) {
                thread_deallocate(old_owner);
        }

        if (old_lingering_owner != THREAD_NULL) {
                thread_deallocate(old_lingering_owner);
        }

        assert(*retval >= 0);

munge_retval:
        if (workq_callback) {
                thread_reenable_sched_call(self, workq_callback);
        }

        if ((flags & ULF_NO_ERRNO) && (ret != 0)) {
                *retval = -ret;
                ret = 0;
        }
        return ret;
}

int
ulock_wake(struct proc *p, struct ulock_wake_args *args, __unused int32_t *retval)
{
        uint opcode = args->operation & UL_OPCODE_MASK;
        uint flags = args->operation & UL_FLAGS_MASK;
        int ret = 0;
        int id = thread_tid(current_thread());
        ulk_t key;

        /* involved threads - each variable holds +1 ref if not null */
        thread_t wake_thread    = THREAD_NULL;
        thread_t old_owner      = THREAD_NULL;

        if (ull_debug) {
                kprintf("[%d]%s>ENTER opcode %d addr %llx flags %x\n",
                        id, __FUNCTION__, opcode, (unsigned long long)(args->addr), flags);
        }

        if ((flags & ULF_WAKE_MASK) != flags) {
                ret = EINVAL;
                goto munge_retval;
        }

#if DEVELOPMENT || DEBUG
        if (opcode == UL_DEBUG_HASH_DUMP_PID) {
                *retval = ull_hash_dump(p->p_pid);
                return ret;
        } else if (opcode == UL_DEBUG_HASH_DUMP_ALL) {
                *retval = ull_hash_dump(0);
                return ret;
        } else if (opcode == UL_DEBUG_SIMULATE_COPYIN_FAULT) {
                ull_simulate_copyin_fault = (int)(args->wake_value);
                return ret;
        }
#endif

        if (args->addr == 0) {
                ret = EINVAL;
                goto munge_retval;
        }

        if (flags & ULF_WAKE_THREAD) {
                if (flags & ULF_WAKE_ALL) {
                        ret = EINVAL;
                        goto munge_retval;
                }
                mach_port_name_t wake_thread_name = (mach_port_name_t)(args->wake_value);
                wake_thread = port_name_to_thread_for_ulock(wake_thread_name);
                if (wake_thread == THREAD_NULL) {
                        ret = ESRCH;
                        goto munge_retval;
                }
        }

        key.ulk_pid = p->p_pid;
        key.ulk_addr = args->addr;

        ull_t *ull = ull_get(&key, ULL_MUST_EXIST);
        if (ull == NULL) {
                if (wake_thread != THREAD_NULL) {
                        thread_deallocate(wake_thread);
                }
                ret = ENOENT;
                goto munge_retval;
        }
        /* ull is locked */

        boolean_t clear_owner = FALSE; /* need to reset owner */

        switch (opcode) {
        case UL_UNFAIR_LOCK:
                clear_owner = TRUE;
                break;
        case UL_COMPARE_AND_WAIT:
                break;
        default:
                if (ull_debug) {
                        kprintf("[%d]%s>EINVAL opcode %d addr 0x%llx flags 0x%x\n",
                                id, __FUNCTION__, opcode, (unsigned long long)(args->addr), flags);
                }
                ret = EINVAL;
                goto out_locked;
        }

        if (opcode != ull->ull_opcode) {
                if (ull_debug) {
                        kprintf("[%d]%s>EDOM - opcode mismatch - opcode %d addr 0x%llx flags 0x%x\n",
                                id, __FUNCTION__, opcode, (unsigned long long)(args->addr), flags);
                }
                ret = EDOM;
                goto out_locked;
        }

        if (!clear_owner) {
                assert(ull->ull_owner == THREAD_NULL);
        }

        if (flags & ULF_WAKE_ALL) {
                thread_wakeup((event_t)ull);
        } else if (flags & ULF_WAKE_THREAD) {
                kern_return_t kr = thread_wakeup_thread((event_t)ull, wake_thread);
                if (kr != KERN_SUCCESS) {
                        assert(kr == KERN_NOT_WAITING);
                        ret = EALREADY;
                }
        } else {
                /*
                 * TODO: WAITQ_SELECT_MAX_PRI forces a linear scan of the (hashed) global waitq.
                 * Move to a ulock-private, priority sorted waitq to avoid that.
                 *
                 * TODO: 'owner is not current_thread (or null)' likely means we can avoid this wakeup
                 * <rdar://problem/25487001>
                 */
                thread_wakeup_one_with_pri((event_t)ull, WAITQ_SELECT_MAX_PRI);
        }

        /*
         * Reaching this point means I previously moved the lock to 'unowned' state in userspace.
         * Therefore I need to relinquish my promotion.
         *
         * However, someone else could have locked it after I unlocked, and then had a third thread
         * block on the lock, causing a promotion of some other owner.
         *
         * I don't want to stomp over that, so only remove the promotion if I'm the current owner.
         */

        if (ull->ull_owner == current_thread()) {
                old_owner = ull_promote_owner_locked(ull, THREAD_NULL);
        }

out_locked:
        ull_put(ull);

        if (wake_thread != THREAD_NULL) {
                thread_deallocate(wake_thread);
        }

        if (old_owner != THREAD_NULL) {
                thread_deallocate(old_owner);
        }

munge_retval:
        if ((flags & ULF_NO_ERRNO) && (ret != 0)) {
                *retval = -ret;
                ret = 0;
        }
        return ret;
}

/*
 * Change ull_owner to be new_owner, and update it with the properties
 * of the current thread.
 *
 * Records the highest current promotion value in ull_promote_token, and applies that
 * to any new owner.
 *
 * Returns +1 ref to the old ull_owner if it is going away.
 */
static thread_t
ull_promote_owner_locked(ull_t*    ull,
                         thread_t  new_owner)
{
        if (new_owner != THREAD_NULL && ull->ull_owner == new_owner) {
                thread_user_promotion_update(new_owner, current_thread(), &ull->ull_promote_token);
                return THREAD_NULL;
        }

        thread_t old_owner = ull->ull_owner;
        ull->ull_owner = THREAD_NULL;

        if (new_owner != THREAD_NULL) {
                /* The ull_owner field now owns a +1 ref on thread */
                thread_reference(new_owner);
                ull->ull_owner = new_owner;

                thread_user_promotion_add(new_owner, current_thread(), &ull->ull_promote_token);
        } else {
                /* No new owner - clear the saturated promotion value */
                ull->ull_promote_token = PROMOTE_TOKEN_INIT;
        }

        if (old_owner != THREAD_NULL) {
                thread_user_promotion_drop(old_owner);
        }

        /* Return the +1 ref from the ull_owner field */
        return old_owner;
}