]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kern/locks.c
xnu-4570.1.46.tar.gz
[apple/xnu.git] / osfmk / kern / locks.c
index 47f39aafbb5a1ffe6b7b44fc8f3dad693deb1ac4..25641b8beb409ef593dec5205996e71db2ae7f05 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2016 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
  * any improvements or extensions that they make and grant Carnegie Mellon
  * the rights to redistribute these changes.
  */
-#include <mach_kdb.h>
+
+#define ATOMIC_PRIVATE 1
+#define LOCK_PRIVATE 1
+
 #include <mach_ldebug.h>
 #include <debug.h>
 
 #include <kern/processor.h>
 #include <kern/sched_prim.h>
 #include <kern/debug.h>
+#include <machine/atomic.h>
+#include <machine/machine_cpu.h>
 #include <string.h>
 
 
 #include <sys/kdebug.h>
 
+#if    CONFIG_DTRACE
+/*
+ * We need only enough declarations from the BSD-side to be able to
+ * test if our probe is active, and to call __dtrace_probe().  Setting
+ * NEED_DTRACE_DEFS gets a local copy of those definitions pulled in.
+ */
+#define NEED_DTRACE_DEFS
+#include <../bsd/sys/lockstat.h>
+#endif
+
 #define        LCK_MTX_SLEEP_CODE              0
 #define        LCK_MTX_SLEEP_DEADLINE_CODE     1
 #define        LCK_MTX_LCK_WAIT_CODE           2
 #define        LCK_MTX_UNLCK_WAKEUP_CODE       3
 
+#if MACH_LDEBUG
+#define ALIGN_TEST(p,t) do{if((uintptr_t)p&(sizeof(t)-1)) __builtin_trap();}while(0)
+#else
+#define ALIGN_TEST(p,t) do{}while(0)
+#endif
+
+/* Silence the volatile to _Atomic cast warning */
+#define ATOMIC_CAST(t,p) ((_Atomic t*)(uintptr_t)(p))
+
+/* Enforce program order of loads and stores. */
+#define ordered_load(target, type) \
+               __c11_atomic_load((_Atomic type *)(target), memory_order_relaxed)
+#define ordered_store(target, type, value) \
+               __c11_atomic_store((_Atomic type *)(target), value, memory_order_relaxed)
+
+#define ordered_load_hw(lock)                  ordered_load(&(lock)->lock_data, uintptr_t)
+#define ordered_store_hw(lock, value)  ordered_store(&(lock)->lock_data, uintptr_t, (value))
+
+#define NOINLINE               __attribute__((noinline))
+
 
 static queue_head_t    lck_grp_queue;
 static unsigned int    lck_grp_cnt;
 
-decl_mutex_data(static,lck_grp_lock)
+decl_lck_mtx_data(static,lck_grp_lock)
+static lck_mtx_ext_t lck_grp_lock_ext;
 
 lck_grp_attr_t LockDefaultGroupAttr;
-lck_grp_t      LockCompatGroup;
-lck_attr_t     LockDefaultLckAttr;
+lck_grp_t              LockCompatGroup;
+lck_attr_t             LockDefaultLckAttr;
+
+#if CONFIG_DTRACE && __SMP__
+#if defined (__x86_64__)
+uint64_t dtrace_spin_threshold = 500; // 500ns
+#elif defined(__arm__) || defined(__arm64__)
+uint64_t dtrace_spin_threshold = LOCK_PANIC_TIMEOUT / 1000000; // 500ns
+#endif
+#endif
 
 /*
  * Routine:    lck_mod_init
@@ -96,12 +140,42 @@ void
 lck_mod_init(
        void)
 {
+       /*
+        * Obtain "lcks" options:this currently controls lock statistics
+        */
+       if (!PE_parse_boot_argn("lcks", &LcksOpts, sizeof (LcksOpts)))
+               LcksOpts = 0;
+
+
+#if (DEVELOPMENT || DEBUG) && defined(__x86_64__)
+       if (!PE_parse_boot_argn("-disable_mtx_chk", &LckDisablePreemptCheck, sizeof (LckDisablePreemptCheck)))
+               LckDisablePreemptCheck = 0;
+#endif /* (DEVELOPMENT || DEBUG) && defined(__x86_64__) */
+
        queue_init(&lck_grp_queue);
-       mutex_init(&lck_grp_lock, 0);
-       lck_grp_cnt = 0;
-       lck_grp_attr_setdefault( &LockDefaultGroupAttr);
-       lck_grp_init( &LockCompatGroup, "Compatibility APIs", LCK_GRP_ATTR_NULL);
+       
+       /* 
+        * Need to bootstrap the LockCompatGroup instead of calling lck_grp_init() here. This avoids
+        * grabbing the lck_grp_lock before it is initialized.
+        */
+       
+       bzero(&LockCompatGroup, sizeof(lck_grp_t));
+       (void) strncpy(LockCompatGroup.lck_grp_name, "Compatibility APIs", LCK_GRP_MAX_NAME);
+       
+       if (LcksOpts & enaLkStat)
+               LockCompatGroup.lck_grp_attr = LCK_GRP_ATTR_STAT;
+    else
+               LockCompatGroup.lck_grp_attr = LCK_ATTR_NONE;
+       
+       LockCompatGroup.lck_grp_refcnt = 1;
+       
+       enqueue_tail(&lck_grp_queue, (queue_entry_t)&LockCompatGroup);
+       lck_grp_cnt = 1;
+       
+       lck_grp_attr_setdefault(&LockDefaultGroupAttr);
        lck_attr_setdefault(&LockDefaultLckAttr);
+       
+       lck_mtx_init_ext(&lck_grp_lock, &lck_grp_lock_ext, &LockCompatGroup, &LockDefaultLckAttr);
 }
 
 /*
@@ -144,7 +218,7 @@ void
 lck_grp_attr_setstat(
        lck_grp_attr_t  *attr)
 {
-       (void)hw_atomic_or((uint32_t *)&attr->grp_attr_val, LCK_GRP_ATTR_STAT);
+       (void)hw_atomic_or(&attr->grp_attr_val, LCK_GRP_ATTR_STAT);
 }
 
 
@@ -161,7 +235,7 @@ lck_grp_attr_free(
 
 
 /*
- * Routine:    lck_grp_alloc_init
+ * Routine: lck_grp_alloc_init
  */
 
 lck_grp_t *
@@ -177,38 +251,35 @@ lck_grp_alloc_init(
        return(grp);
 }
 
-
 /*
- * Routine:    lck_grp_init
+ * Routine: lck_grp_init
  */
 
 void
-lck_grp_init(
-       lck_grp_t               *grp,               
-       const char*             grp_name,           
-       lck_grp_attr_t  *attr)             
+lck_grp_init(lck_grp_t * grp, const char * grp_name, lck_grp_attr_t * attr)
 {
+       /* make sure locking infrastructure has been initialized */
+       assert(lck_grp_cnt > 0);
+
        bzero((void *)grp, sizeof(lck_grp_t));
 
-       (void) strncpy(grp->lck_grp_name, grp_name, LCK_GRP_MAX_NAME);
+       (void)strlcpy(grp->lck_grp_name, grp_name, LCK_GRP_MAX_NAME);
 
        if (attr != LCK_GRP_ATTR_NULL)
                grp->lck_grp_attr = attr->grp_attr_val;
        else if (LcksOpts & enaLkStat)
-                grp->lck_grp_attr = LCK_GRP_ATTR_STAT;
-        else
-                grp->lck_grp_attr = LCK_ATTR_NONE;
+               grp->lck_grp_attr = LCK_GRP_ATTR_STAT;
+       else
+               grp->lck_grp_attr = LCK_ATTR_NONE;
 
        grp->lck_grp_refcnt = 1;
 
-       mutex_lock(&lck_grp_lock);
+       lck_mtx_lock(&lck_grp_lock);
        enqueue_tail(&lck_grp_queue, (queue_entry_t)grp);
        lck_grp_cnt++;
-       mutex_unlock(&lck_grp_lock);
-
+       lck_mtx_unlock(&lck_grp_lock);
 }
 
-
 /*
  * Routine:    lck_grp_free
  */
@@ -217,10 +288,10 @@ void
 lck_grp_free(
        lck_grp_t       *grp)
 {
-       mutex_lock(&lck_grp_lock);
+       lck_mtx_lock(&lck_grp_lock);
        lck_grp_cnt--;
        (void)remque((queue_entry_t)grp);
-       mutex_unlock(&lck_grp_lock);
+       lck_mtx_unlock(&lck_grp_lock);
        lck_grp_deallocate(grp);
 }
 
@@ -233,7 +304,7 @@ void
 lck_grp_reference(
        lck_grp_t       *grp)
 {
-       (void)hw_atomic_add((uint32_t *)(&grp->lck_grp_refcnt), 1);
+       (void)hw_atomic_add(&grp->lck_grp_refcnt, 1);
 }
 
 
@@ -245,7 +316,7 @@ void
 lck_grp_deallocate(
        lck_grp_t       *grp)
 {
-       if (hw_atomic_sub((uint32_t *)(&grp->lck_grp_refcnt), 1) == 0)
+       if (hw_atomic_sub(&grp->lck_grp_refcnt, 1) == 0)
                kfree(grp, sizeof(lck_grp_t));
 }
 
@@ -274,7 +345,7 @@ lck_grp_lckcnt_incr(
                return panic("lck_grp_lckcnt_incr(): invalid lock type: %d\n", lck_type);
        }
 
-       (void)hw_atomic_add((uint32_t *)lckcnt, 1);
+       (void)hw_atomic_add(lckcnt, 1);
 }
 
 /*
@@ -287,6 +358,7 @@ lck_grp_lckcnt_decr(
        lck_type_t      lck_type)
 {
        unsigned int    *lckcnt;
+       int             updated;
 
        switch (lck_type) {
        case LCK_TYPE_SPIN:
@@ -299,10 +371,12 @@ lck_grp_lckcnt_decr(
                lckcnt = &grp->lck_grp_rwcnt;
                break;
        default:
-               return panic("lck_grp_lckcnt_decr(): invalid lock type: %d\n", lck_type);
+               panic("lck_grp_lckcnt_decr(): invalid lock type: %d\n", lck_type);
+               return;
        }
 
-       (void)hw_atomic_sub((uint32_t *)lckcnt, 1);
+       updated = (int)hw_atomic_sub(lckcnt, 1);
+       assert(updated >= 0);
 }
 
 /*
@@ -330,15 +404,21 @@ void
 lck_attr_setdefault(
        lck_attr_t      *attr)
 {
+#if __arm__ || __arm64__
+       /* <rdar://problem/4404579>: Using LCK_ATTR_DEBUG here causes panic at boot time for arm */
+       attr->lck_attr_val =  LCK_ATTR_NONE;
+#elif __i386__ || __x86_64__
 #if     !DEBUG
-       if (LcksOpts & enaLkDeb)
-               attr->lck_attr_val =  LCK_ATTR_DEBUG;
-       else
-               attr->lck_attr_val =  LCK_ATTR_NONE;
+       if (LcksOpts & enaLkDeb)
+               attr->lck_attr_val =  LCK_ATTR_DEBUG;
+       else
+               attr->lck_attr_val =  LCK_ATTR_NONE;
 #else
-       attr->lck_attr_val =  LCK_ATTR_DEBUG;
-#endif
-
+       attr->lck_attr_val =  LCK_ATTR_DEBUG;
+#endif /* !DEBUG */
+#else
+#error Unknown architecture.
+#endif /* __arm__ */
 }
 
 
@@ -349,7 +429,17 @@ void
 lck_attr_setdebug(
        lck_attr_t      *attr)
 {
-       (void)hw_atomic_or((uint32_t *)&attr->lck_attr_val, LCK_ATTR_DEBUG);
+       (void)hw_atomic_or(&attr->lck_attr_val, LCK_ATTR_DEBUG);
+}
+
+/*
+ * Routine:    lck_attr_setdebug
+ */
+void
+lck_attr_cleardebug(
+       lck_attr_t      *attr)
+{
+       (void)hw_atomic_and(&attr->lck_attr_val, ~LCK_ATTR_DEBUG);
 }
 
 
@@ -360,7 +450,7 @@ void
 lck_attr_rw_shared_priority(
        lck_attr_t      *attr)
 {
-       (void)hw_atomic_or((uint32_t *)&attr->lck_attr_val, LCK_ATTR_RW_SHARED_PRIORITY);
+       (void)hw_atomic_or(&attr->lck_attr_val, LCK_ATTR_RW_SHARED_PRIORITY);
 }
 
 
@@ -374,6 +464,246 @@ lck_attr_free(
        kfree(attr, sizeof(lck_attr_t));
 }
 
+/*
+ * Routine:    hw_lock_init
+ *
+ *     Initialize a hardware lock.
+ */
+void
+hw_lock_init(hw_lock_t lock)
+{
+       ordered_store_hw(lock, 0);
+}
+
+/*
+ *     Routine: hw_lock_lock_contended
+ *
+ *     Spin until lock is acquired or timeout expires.
+ *     timeout is in mach_absolute_time ticks. Called with
+ *     preemption disabled.
+ */
+
+#if    __SMP__
+static unsigned int NOINLINE
+hw_lock_lock_contended(hw_lock_t lock, uintptr_t data, uint64_t timeout, boolean_t do_panic)
+{
+       uint64_t        end = 0;
+       uintptr_t       holder = lock->lock_data;
+       int             i;
+
+       if (timeout == 0)
+               timeout = LOCK_PANIC_TIMEOUT;
+#if CONFIG_DTRACE
+       uint64_t begin;
+       boolean_t dtrace_enabled = lockstat_probemap[LS_LCK_SPIN_LOCK_SPIN] != 0;
+       if (__improbable(dtrace_enabled))
+               begin = mach_absolute_time();
+#endif
+       for ( ; ; ) {   
+               for (i = 0; i < LOCK_SNOOP_SPINS; i++) {
+                       cpu_pause();
+#if (!__ARM_ENABLE_WFE_) || (LOCK_PRETEST)
+                       holder = ordered_load_hw(lock);
+                       if (holder != 0)
+                               continue;
+#endif
+                       if (atomic_compare_exchange(&lock->lock_data, 0, data,
+                           memory_order_acquire_smp, TRUE)) {
+#if CONFIG_DTRACE
+                               if (__improbable(dtrace_enabled)) {
+                                       uint64_t spintime = mach_absolute_time() - begin;
+                                       if (spintime > dtrace_spin_threshold)
+                                               LOCKSTAT_RECORD2(LS_LCK_SPIN_LOCK_SPIN, lock, spintime, dtrace_spin_threshold);
+                               }
+#endif
+                               return 1;
+                       }
+               }
+               if (end == 0) {
+                       end = ml_get_timebase() + timeout;
+               }
+               else if (ml_get_timebase() >= end)
+                       break;
+       }
+       if (do_panic) {
+               // Capture the actual time spent blocked, which may be higher than the timeout
+               // if a misbehaving interrupt stole this thread's CPU time.
+               panic("Spinlock timeout after %llu ticks, %p = %lx",
+                       (ml_get_timebase() - end + timeout), lock, holder);
+       }
+       return 0;
+}
+#endif // __SMP__
+
+/*
+ *     Routine: hw_lock_lock
+ *
+ *     Acquire lock, spinning until it becomes available,
+ *     return with preemption disabled.
+ */
+void
+hw_lock_lock(hw_lock_t lock)
+{
+       thread_t        thread;
+       uintptr_t       state;
+
+       thread = current_thread();
+       disable_preemption_for_thread(thread);
+       state = LCK_MTX_THREAD_TO_STATE(thread) | PLATFORM_LCK_ILOCK;
+#if    __SMP__
+
+#if    LOCK_PRETEST
+       if (ordered_load_hw(lock))
+               goto contended;
+#endif // LOCK_PRETEST
+       if (atomic_compare_exchange(&lock->lock_data, 0, state,
+                                       memory_order_acquire_smp, TRUE)) {
+               goto end;
+       }
+#if    LOCK_PRETEST
+contended:
+#endif // LOCK_PRETEST
+       hw_lock_lock_contended(lock, state, 0, TRUE);
+end:
+#else  // __SMP__
+       if (lock->lock_data)
+               panic("Spinlock held %p", lock);
+       lock->lock_data = state;
+#endif // __SMP__
+#if CONFIG_DTRACE
+       LOCKSTAT_RECORD(LS_LCK_SPIN_LOCK_ACQUIRE, lock, 0);
+#endif
+       return;
+}
+
+/*
+ *     Routine: hw_lock_to
+ *
+ *     Acquire lock, spinning until it becomes available or timeout.
+ *     Timeout is in mach_absolute_time ticks, return with
+ *     preemption disabled.
+ */
+unsigned int
+hw_lock_to(hw_lock_t lock, uint64_t timeout)
+{
+       thread_t        thread;
+       uintptr_t       state;
+       unsigned int success = 0;
+
+       thread = current_thread();
+       disable_preemption_for_thread(thread);
+       state = LCK_MTX_THREAD_TO_STATE(thread) | PLATFORM_LCK_ILOCK;
+#if    __SMP__
+
+#if    LOCK_PRETEST
+       if (ordered_load_hw(lock))
+               goto contended;
+#endif // LOCK_PRETEST
+       if (atomic_compare_exchange(&lock->lock_data, 0, state,
+                                       memory_order_acquire_smp, TRUE)) {
+               success = 1;
+               goto end;
+       }
+#if    LOCK_PRETEST
+contended:
+#endif // LOCK_PRETEST
+       success = hw_lock_lock_contended(lock, state, timeout, FALSE);
+end:
+#else  // __SMP__
+       (void)timeout;
+       if (ordered_load_hw(lock) == 0) {
+               ordered_store_hw(lock, state);
+               success = 1;
+       }
+#endif // __SMP__
+#if CONFIG_DTRACE
+       if (success)
+               LOCKSTAT_RECORD(LS_LCK_SPIN_LOCK_ACQUIRE, lock, 0);
+#endif
+       return success;
+}
+
+/*
+ *     Routine: hw_lock_try
+ *
+ *     returns with preemption disabled on success.
+ */
+unsigned int
+hw_lock_try(hw_lock_t lock)
+{
+       thread_t        thread = current_thread();
+       int             success = 0;
+#if    LOCK_TRY_DISABLE_INT
+       long            intmask;
+
+       intmask = disable_interrupts();
+#else
+       disable_preemption_for_thread(thread);
+#endif // LOCK_TRY_DISABLE_INT
+
+#if    __SMP__
+#if    LOCK_PRETEST
+       if (ordered_load_hw(lock))
+               goto failed;
+#endif // LOCK_PRETEST
+       success = atomic_compare_exchange(&lock->lock_data, 0, LCK_MTX_THREAD_TO_STATE(thread) | PLATFORM_LCK_ILOCK,
+                                       memory_order_acquire_smp, FALSE);
+#else
+       if (lock->lock_data == 0) {
+               lock->lock_data = LCK_MTX_THREAD_TO_STATE(thread) | PLATFORM_LCK_ILOCK;
+               success = 1;
+       }
+#endif // __SMP__
+
+#if    LOCK_TRY_DISABLE_INT
+       if (success)
+               disable_preemption_for_thread(thread);
+#if    LOCK_PRETEST
+failed:
+#endif // LOCK_PRETEST
+       restore_interrupts(intmask);
+#else
+#if    LOCK_PRETEST
+failed:
+#endif // LOCK_PRETEST
+       if (!success)
+               enable_preemption();
+#endif // LOCK_TRY_DISABLE_INT
+#if CONFIG_DTRACE
+       if (success)
+               LOCKSTAT_RECORD(LS_LCK_SPIN_LOCK_ACQUIRE, lock, 0);
+#endif
+       return success;
+}
+
+/*
+ *     Routine: hw_lock_unlock
+ *
+ *     Unconditionally release lock, release preemption level.
+ */
+void
+hw_lock_unlock(hw_lock_t lock)
+{
+       __c11_atomic_store((_Atomic uintptr_t *)&lock->lock_data, 0, memory_order_release_smp);
+#if __arm__ || __arm64__
+       // ARM tests are only for open-source exclusion
+       set_event();
+#endif // __arm__ || __arm64__
+#if    CONFIG_DTRACE
+       LOCKSTAT_RECORD(LS_LCK_SPIN_UNLOCK_RELEASE, lock, 0);
+#endif /* CONFIG_DTRACE */
+       enable_preemption();
+}
+
+/*
+ *     Routine hw_lock_held, doesn't change preemption state.
+ *     N.B.  Racy, of course.
+ */
+unsigned int
+hw_lock_held(hw_lock_t lock)
+{
+       return (ordered_load_hw(lock) != 0);
+}
 
 /*
  * Routine:    lck_spin_sleep
@@ -436,6 +766,39 @@ lck_spin_sleep_deadline(
 }
 
 
+/*
+ * Routine:    lck_mtx_clear_promoted
+ *
+ * Handle clearing of TH_SFLAG_PROMOTED,
+ * adjusting thread priority as needed.
+ *
+ * Called with thread lock held
+ */
+static void
+lck_mtx_clear_promoted (
+       thread_t                        thread,
+       __kdebug_only uintptr_t         trace_lck)
+{
+       thread->sched_flags &= ~TH_SFLAG_PROMOTED;
+
+       if (thread->sched_flags & TH_SFLAG_RW_PROMOTED) {
+               /* Thread still has a RW lock promotion */
+       } else if (thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) {
+               KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED,MACH_DEMOTE) | DBG_FUNC_NONE,
+                               thread->sched_pri, DEPRESSPRI, 0, trace_lck, 0);
+               set_sched_pri(thread, DEPRESSPRI);
+       } else {
+               if (thread->base_pri < thread->sched_pri) {
+                       KERNEL_DEBUG_CONSTANT(
+                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_DEMOTE) | DBG_FUNC_NONE,
+                                       thread->sched_pri, thread->base_pri, 0, trace_lck, 0);
+               }
+               thread_recompute_sched_pri(thread, FALSE);
+       }
+}
+
+
 /*
  * Routine:    lck_mtx_sleep
  */
@@ -447,24 +810,48 @@ lck_mtx_sleep(
        wait_interrupt_t        interruptible)
 {
        wait_result_t   res;
+       thread_t                thread = current_thread();
  
        KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_SLEEP_CODE) | DBG_FUNC_START,
-                    (int)lck, (int)lck_sleep_action, (int)event, (int)interruptible, 0);
+                    VM_KERNEL_UNSLIDE_OR_PERM(lck), (int)lck_sleep_action, VM_KERNEL_UNSLIDE_OR_PERM(event), (int)interruptible, 0);
 
        if ((lck_sleep_action & ~LCK_SLEEP_MASK) != 0)
                panic("Invalid lock sleep action %x\n", lck_sleep_action);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               /*
+                * We overload the RW lock promotion to give us a priority ceiling
+                * during the time that this thread is asleep, so that when it
+                * is re-awakened (and not yet contending on the mutex), it is
+                * runnable at a reasonably high priority.
+                */
+               thread->rwlock_count++;
+       }
+
        res = assert_wait(event, interruptible);
        if (res == THREAD_WAITING) {
                lck_mtx_unlock(lck);
                res = thread_block(THREAD_CONTINUE_NULL);
-               if (!(lck_sleep_action & LCK_SLEEP_UNLOCK))
-                       lck_mtx_lock(lck);
+               if (!(lck_sleep_action & LCK_SLEEP_UNLOCK)) {
+                       if ((lck_sleep_action & LCK_SLEEP_SPIN))
+                               lck_mtx_lock_spin(lck);
+                       else if ((lck_sleep_action & LCK_SLEEP_SPIN_ALWAYS))
+                               lck_mtx_lock_spin_always(lck);
+                       else
+                               lck_mtx_lock(lck);
+               }
        }
        else
        if (lck_sleep_action & LCK_SLEEP_UNLOCK)
                lck_mtx_unlock(lck);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               if ((thread->rwlock_count-- == 1 /* field now 0 */) && (thread->sched_flags & TH_SFLAG_RW_PROMOTED)) {
+                       /* sched_flags checked without lock, but will be rechecked while clearing */
+                       lck_rw_clear_promotion(thread);
+               }
+       }
+
        KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_SLEEP_CODE) | DBG_FUNC_END, (int)res, 0, 0, 0, 0);
 
        return res;
@@ -483,24 +870,43 @@ lck_mtx_sleep_deadline(
        uint64_t                deadline)
 {
        wait_result_t   res;
+       thread_t                thread = current_thread();
 
        KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_SLEEP_DEADLINE_CODE) | DBG_FUNC_START,
-                    (int)lck, (int)lck_sleep_action, (int)event, (int)interruptible, 0);
+                    VM_KERNEL_UNSLIDE_OR_PERM(lck), (int)lck_sleep_action, VM_KERNEL_UNSLIDE_OR_PERM(event), (int)interruptible, 0);
 
        if ((lck_sleep_action & ~LCK_SLEEP_MASK) != 0)
                panic("Invalid lock sleep action %x\n", lck_sleep_action);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               /*
+                * See lck_mtx_sleep().
+                */
+               thread->rwlock_count++;
+       }
+
        res = assert_wait_deadline(event, interruptible, deadline);
        if (res == THREAD_WAITING) {
                lck_mtx_unlock(lck);
                res = thread_block(THREAD_CONTINUE_NULL);
-               if (!(lck_sleep_action & LCK_SLEEP_UNLOCK))
-                       lck_mtx_lock(lck);
+               if (!(lck_sleep_action & LCK_SLEEP_UNLOCK)) {
+                       if ((lck_sleep_action & LCK_SLEEP_SPIN))
+                               lck_mtx_lock_spin(lck);
+                       else
+                               lck_mtx_lock(lck);
+               }
        }
        else
        if (lck_sleep_action & LCK_SLEEP_UNLOCK)
                lck_mtx_unlock(lck);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               if ((thread->rwlock_count-- == 1 /* field now 0 */) && (thread->sched_flags & TH_SFLAG_RW_PROMOTED)) {
+                       /* sched_flags checked without lock, but will be rechecked while clearing */
+                       lck_rw_clear_promotion(thread);
+               }
+       }
+
        KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_SLEEP_DEADLINE_CODE) | DBG_FUNC_END, (int)res, 0, 0, 0, 0);
 
        return res;
@@ -521,38 +927,45 @@ lck_mtx_lock_wait (
 {
        thread_t                self = current_thread();
        lck_mtx_t               *mutex;
+       __kdebug_only uintptr_t trace_lck = VM_KERNEL_UNSLIDE_OR_PERM(lck);
+       __kdebug_only uintptr_t trace_holder = VM_KERNEL_UNSLIDE_OR_PERM(holder);
        integer_t               priority;
        spl_t                   s = splsched();
+#if    CONFIG_DTRACE
+       uint64_t                sleep_start = 0;
+
+       if (lockstat_probemap[LS_LCK_MTX_LOCK_BLOCK] || lockstat_probemap[LS_LCK_MTX_EXT_LOCK_BLOCK]) {
+               sleep_start = mach_absolute_time();
+       }
+#endif
 
        if (lck->lck_mtx_tag != LCK_MTX_TAG_INDIRECT)
                mutex = lck;
        else
                mutex = &lck->lck_mtx_ptr->lck_mtx;
 
-       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_LCK_WAIT_CODE) | DBG_FUNC_START, (int)lck, (int)holder, 0, 0, 0);
+       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_LCK_WAIT_CODE) | DBG_FUNC_START, trace_lck, trace_holder, 0, 0, 0);
 
        priority = self->sched_pri;
-       if (priority < self->priority)
-               priority = self->priority;
-       if (priority > MINPRI_KERNEL)
-               priority = MINPRI_KERNEL;
-       else
+       if (priority < self->base_pri)
+               priority = self->base_pri;
        if (priority < BASEPRI_DEFAULT)
                priority = BASEPRI_DEFAULT;
 
+       /* Do not promote past promotion ceiling */
+       priority = MIN(priority, MAXPRI_PROMOTE);
+
        thread_lock(holder);
-       if (mutex->lck_mtx_pri == 0)
+       if (mutex->lck_mtx_pri == 0) {
                holder->promotions++;
-       if (holder->priority < MINPRI_KERNEL) {
-               holder->sched_mode |= TH_MODE_PROMOTED;
-               if (    mutex->lck_mtx_pri < priority   &&
-                               holder->sched_pri < priority            ) {
-                       KERNEL_DEBUG_CONSTANT(
-                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_PROMOTE) | DBG_FUNC_NONE,
-                                       holder->sched_pri, priority, (int)holder, (int)lck, 0);
+               holder->sched_flags |= TH_SFLAG_PROMOTED;
+       }
 
-                       set_sched_pri(holder, priority);
-               }
+       if (mutex->lck_mtx_pri < priority && holder->sched_pri < priority) {
+               KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED,MACH_PROMOTE) | DBG_FUNC_NONE,
+                                       holder->sched_pri, priority, trace_holder, trace_lck, 0);
+               set_sched_pri(holder, priority);
        }
        thread_unlock(holder);
        splx(s);
@@ -569,12 +982,28 @@ lck_mtx_lock_wait (
                mutex->lck_mtx_waiters++;
        }
 
-       assert_wait((event_t)(((unsigned int*)lck)+((sizeof(lck_mtx_t)-1)/sizeof(unsigned int))), THREAD_UNINT);
+       thread_set_pending_block_hint(self, kThreadWaitKernelMutex);
+       assert_wait(LCK_MTX_EVENT(mutex), THREAD_UNINT);
        lck_mtx_ilk_unlock(mutex);
 
        thread_block(THREAD_CONTINUE_NULL);
 
        KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_LCK_WAIT_CODE) | DBG_FUNC_END, 0, 0, 0, 0, 0);
+#if    CONFIG_DTRACE
+       /*
+        * Record the DTrace lockstat probe for blocking, block time
+        * measured from when we were entered.
+        */
+       if (sleep_start) {
+               if (lck->lck_mtx_tag != LCK_MTX_TAG_INDIRECT) {
+                       LOCKSTAT_RECORD(LS_LCK_MTX_LOCK_BLOCK, lck,
+                           mach_absolute_time() - sleep_start);
+               } else {
+                       LOCKSTAT_RECORD(LS_LCK_MTX_EXT_LOCK_BLOCK, lck,
+                           mach_absolute_time() - sleep_start);
+               }
+       }
+#endif
 }
 
 /*
@@ -593,6 +1022,9 @@ lck_mtx_lock_acquire(
 {
        thread_t                thread = current_thread();
        lck_mtx_t               *mutex;
+       integer_t               priority;
+       spl_t                   s;
+       __kdebug_only uintptr_t trace_lck = VM_KERNEL_UNSLIDE_OR_PERM(lck);
 
        if (lck->lck_mtx_tag != LCK_MTX_TAG_INDIRECT)
                mutex = lck;
@@ -606,28 +1038,48 @@ lck_mtx_lock_acquire(
                mutex->lck_mtx_waiters--;
        }
 
-       if (mutex->lck_mtx_waiters > 0) {
-               integer_t               priority = mutex->lck_mtx_pri;
-               spl_t                   s = splsched();
+       if (mutex->lck_mtx_waiters)
+               priority = mutex->lck_mtx_pri;
+       else {
+               mutex->lck_mtx_pri = 0;
+               priority = 0;
+       }
 
+       if (priority || thread->was_promoted_on_wakeup) {
+               s = splsched();
                thread_lock(thread);
-               thread->promotions++;
-               if (thread->priority < MINPRI_KERNEL) {
-                       thread->sched_mode |= TH_MODE_PROMOTED;
+
+               if (priority) {
+                       thread->promotions++;
+                       thread->sched_flags |= TH_SFLAG_PROMOTED;
                        if (thread->sched_pri < priority) {
                                KERNEL_DEBUG_CONSTANT(
                                        MACHDBG_CODE(DBG_MACH_SCHED,MACH_PROMOTE) | DBG_FUNC_NONE,
-                                               thread->sched_pri, priority, 0, (int)lck, 0);
-
+                                                       thread->sched_pri, priority, 0, trace_lck, 0);
+                               /* Do not promote past promotion ceiling */
+                               assert(priority <= MAXPRI_PROMOTE);
                                set_sched_pri(thread, priority);
                        }
                }
+               if (thread->was_promoted_on_wakeup) {
+                       thread->was_promoted_on_wakeup = 0;
+                       if (thread->promotions == 0)
+                               lck_mtx_clear_promoted(thread, trace_lck);
+               }
+
                thread_unlock(thread);
                splx(s);
        }
-       else
-               mutex->lck_mtx_pri = 0;
 
+#if CONFIG_DTRACE
+       if (lockstat_probemap[LS_LCK_MTX_LOCK_ACQUIRE] || lockstat_probemap[LS_LCK_MTX_EXT_LOCK_ACQUIRE]) {
+               if (lck->lck_mtx_tag != LCK_MTX_TAG_INDIRECT) {
+                       LOCKSTAT_RECORD(LS_LCK_MTX_LOCK_ACQUIRE, lck, 0);
+               } else {
+                       LOCKSTAT_RECORD(LS_LCK_MTX_EXT_LOCK_ACQUIRE, lck, 0);
+               }
+       }
+#endif 
        return (mutex->lck_mtx_waiters);
 }
 
@@ -645,71 +1097,121 @@ lck_mtx_unlock_wakeup (
 {
        thread_t                thread = current_thread();
        lck_mtx_t               *mutex;
+       __kdebug_only uintptr_t trace_lck = VM_KERNEL_UNSLIDE_OR_PERM(lck);
 
        if (lck->lck_mtx_tag != LCK_MTX_TAG_INDIRECT)
                mutex = lck;
        else
                mutex = &lck->lck_mtx_ptr->lck_mtx;
 
+       if (thread != holder)
+               panic("lck_mtx_unlock_wakeup: mutex %p holder %p\n", mutex, holder);
 
-       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_UNLCK_WAKEUP_CODE) | DBG_FUNC_START, (int)lck, (int)holder, 0, 0, 0);
+       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_UNLCK_WAKEUP_CODE) | DBG_FUNC_START, trace_lck, VM_KERNEL_UNSLIDE_OR_PERM(holder), 0, 0, 0);
 
-       if (thread != holder)
-               panic("lck_mtx_unlock_wakeup: mutex %x holder %x\n", mutex, holder);
+       assert(mutex->lck_mtx_waiters > 0);
+       if (mutex->lck_mtx_waiters > 1)
+               thread_wakeup_one_with_pri(LCK_MTX_EVENT(lck), lck->lck_mtx_pri);
+       else
+               thread_wakeup_one(LCK_MTX_EVENT(lck));
 
        if (thread->promotions > 0) {
                spl_t           s = splsched();
 
                thread_lock(thread);
-               if (    --thread->promotions == 0                               &&
-                               (thread->sched_mode & TH_MODE_PROMOTED)         ) {
-                       thread->sched_mode &= ~TH_MODE_PROMOTED;
-                       if (thread->sched_mode & TH_MODE_ISDEPRESSED) {
-                               KERNEL_DEBUG_CONSTANT(
-                                       MACHDBG_CODE(DBG_MACH_SCHED,MACH_DEMOTE) | DBG_FUNC_NONE,
-                                                 thread->sched_pri, DEPRESSPRI, 0, (int)lck, 0);
-
-                               set_sched_pri(thread, DEPRESSPRI);
-                       }
-                       else {
-                               if (thread->priority < thread->sched_pri) {
-                                       KERNEL_DEBUG_CONSTANT(
-                                               MACHDBG_CODE(DBG_MACH_SCHED,MACH_DEMOTE) |
-                                                                                                                       DBG_FUNC_NONE,
-                                                       thread->sched_pri, thread->priority,
-                                                                       0, (int)lck, 0);
-                               }
-
-                               compute_priority(thread, FALSE);
-                       }
-               }
+               if (--thread->promotions == 0 && (thread->sched_flags & TH_SFLAG_PROMOTED))
+                       lck_mtx_clear_promoted(thread, trace_lck);
                thread_unlock(thread);
                splx(s);
        }
-       assert(mutex->lck_mtx_waiters > 0);
-       thread_wakeup_one((event_t)(((unsigned int*)lck)+(sizeof(lck_mtx_t)-1)/sizeof(unsigned int)));
 
        KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_UNLCK_WAKEUP_CODE) | DBG_FUNC_END, 0, 0, 0, 0, 0);
 }
 
+void
+lck_mtx_unlockspin_wakeup (
+       lck_mtx_t                       *lck)
+{
+       assert(lck->lck_mtx_waiters > 0);
+       thread_wakeup_one(LCK_MTX_EVENT(lck));
+
+       KERNEL_DEBUG(MACHDBG_CODE(DBG_MACH_LOCKS, LCK_MTX_UNLCK_WAKEUP_CODE) | DBG_FUNC_NONE, VM_KERNEL_UNSLIDE_OR_PERM(lck), 0, 0, 1, 0);
+#if CONFIG_DTRACE
+       /*
+        * When there are waiters, we skip the hot-patch spot in the
+        * fastpath, so we record it here.
+        */
+       LOCKSTAT_RECORD(LS_LCK_MTX_UNLOCK_RELEASE, lck, 0);
+#endif
+}
+
+
 /*
  * Routine:    mutex_pause
  *
  * Called by former callers of simple_lock_pause().
  */
+#define MAX_COLLISION_COUNTS   32
+#define MAX_COLLISION  8
+
+unsigned int max_collision_count[MAX_COLLISION_COUNTS];
+
+uint32_t collision_backoffs[MAX_COLLISION] = {
+        10, 50, 100, 200, 400, 600, 800, 1000
+};
+
 
 void
-mutex_pause(void)
+mutex_pause(uint32_t collisions)
 {
        wait_result_t wait_result;
+       uint32_t        back_off;
+
+       if (collisions >= MAX_COLLISION_COUNTS)
+               collisions = MAX_COLLISION_COUNTS - 1;
+       max_collision_count[collisions]++;
 
-       wait_result = assert_wait_timeout((event_t)mutex_pause, THREAD_UNINT, 1, 1000*NSEC_PER_USEC);
+       if (collisions >= MAX_COLLISION)
+               collisions = MAX_COLLISION - 1;
+       back_off = collision_backoffs[collisions];
+
+       wait_result = assert_wait_timeout((event_t)mutex_pause, THREAD_UNINT, back_off, NSEC_PER_USEC);
        assert(wait_result == THREAD_WAITING);
 
        wait_result = thread_block(THREAD_CONTINUE_NULL);
        assert(wait_result == THREAD_TIMED_OUT);
 }
 
+
+unsigned int mutex_yield_wait = 0;
+unsigned int mutex_yield_no_wait = 0;
+
+void
+lck_mtx_yield(
+           lck_mtx_t   *lck)
+{
+       int     waiters;
+       
+#if DEBUG
+       lck_mtx_assert(lck, LCK_MTX_ASSERT_OWNED);
+#endif /* DEBUG */
+       
+       if (lck->lck_mtx_tag == LCK_MTX_TAG_INDIRECT)
+               waiters = lck->lck_mtx_ptr->lck_mtx.lck_mtx_waiters;
+       else
+               waiters = lck->lck_mtx_waiters;
+
+       if ( !waiters) {
+               mutex_yield_no_wait++;
+       } else {
+               mutex_yield_wait++;
+               lck_mtx_unlock(lck);
+               mutex_pause(0);
+               lck_mtx_lock(lck);
+       }
+}
+
+
 /*
  * Routine:    lck_rw_sleep
  */
@@ -722,10 +1224,23 @@ lck_rw_sleep(
 {
        wait_result_t   res;
        lck_rw_type_t   lck_rw_type;
+       thread_t                thread = current_thread();
+
        if ((lck_sleep_action & ~LCK_SLEEP_MASK) != 0)
                panic("Invalid lock sleep action %x\n", lck_sleep_action);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               /*
+                * Although we are dropping the RW lock, the intent in most cases
+                * is that this thread remains as an observer, since it may hold
+                * some secondary resource, but must yield to avoid deadlock. In
+                * this situation, make sure that the thread is boosted to the
+                * RW lock ceiling while blocked, so that it can re-acquire the
+                * RW lock at that priority.
+                */
+               thread->rwlock_count++;
+       }
+
        res = assert_wait(event, interruptible);
        if (res == THREAD_WAITING) {
                lck_rw_type = lck_rw_done(lck);
@@ -743,6 +1258,17 @@ lck_rw_sleep(
        if (lck_sleep_action & LCK_SLEEP_UNLOCK)
                (void)lck_rw_done(lck);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               if ((thread->rwlock_count-- == 1 /* field now 0 */) && (thread->sched_flags & TH_SFLAG_RW_PROMOTED)) {
+                       /* sched_flags checked without lock, but will be rechecked while clearing */
+
+                       /* Only if the caller wanted the lck_rw_t returned unlocked should we drop to 0 */
+                       assert(lck_sleep_action & LCK_SLEEP_UNLOCK);
+
+                       lck_rw_clear_promotion(thread);
+               }
+       }
+
        return res;
 }
 
@@ -760,10 +1286,15 @@ lck_rw_sleep_deadline(
 {
        wait_result_t   res;
        lck_rw_type_t   lck_rw_type;
+       thread_t                thread = current_thread();
 
        if ((lck_sleep_action & ~LCK_SLEEP_MASK) != 0)
                panic("Invalid lock sleep action %x\n", lck_sleep_action);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               thread->rwlock_count++;
+       }
+
        res = assert_wait_deadline(event, interruptible, deadline);
        if (res == THREAD_WAITING) {
                lck_rw_type = lck_rw_done(lck);
@@ -781,9 +1312,141 @@ lck_rw_sleep_deadline(
        if (lck_sleep_action & LCK_SLEEP_UNLOCK)
                (void)lck_rw_done(lck);
 
+       if (lck_sleep_action & LCK_SLEEP_PROMOTED_PRI) {
+               if ((thread->rwlock_count-- == 1 /* field now 0 */) && (thread->sched_flags & TH_SFLAG_RW_PROMOTED)) {
+                       /* sched_flags checked without lock, but will be rechecked while clearing */
+
+                       /* Only if the caller wanted the lck_rw_t returned unlocked should we drop to 0 */
+                       assert(lck_sleep_action & LCK_SLEEP_UNLOCK);
+
+                       lck_rw_clear_promotion(thread);
+               }
+       }
+
        return res;
 }
 
+/*
+ * Reader-writer lock promotion
+ *
+ * We support a limited form of reader-writer
+ * lock promotion whose effects are:
+ * 
+ *   * Qualifying threads have decay disabled
+ *   * Scheduler priority is reset to a floor of
+ *     of their statically assigned priority
+ *     or BASEPRI_BACKGROUND
+ *
+ * The rationale is that lck_rw_ts do not have
+ * a single owner, so we cannot apply a directed
+ * priority boost from all waiting threads
+ * to all holding threads without maintaining
+ * lists of all shared owners and all waiting
+ * threads for every lock.
+ *
+ * Instead (and to preserve the uncontended fast-
+ * path), acquiring (or attempting to acquire)
+ * a RW lock in shared or exclusive lock increments
+ * a per-thread counter. Only if that thread stops
+ * making forward progress (for instance blocking
+ * on a mutex, or being preempted) do we consult
+ * the counter and apply the priority floor.
+ * When the thread becomes runnable again (or in
+ * the case of preemption it never stopped being
+ * runnable), it has the priority boost and should
+ * be in a good position to run on the CPU and
+ * release all RW locks (at which point the priority
+ * boost is cleared).
+ *
+ * Care must be taken to ensure that priority
+ * boosts are not retained indefinitely, since unlike
+ * mutex priority boosts (where the boost is tied
+ * to the mutex lifecycle), the boost is tied
+ * to the thread and independent of any particular
+ * lck_rw_t. Assertions are in place on return
+ * to userspace so that the boost is not held
+ * indefinitely.
+ *
+ * The routines that increment/decrement the
+ * per-thread counter should err on the side of
+ * incrementing any time a preemption is possible
+ * and the lock would be visible to the rest of the
+ * system as held (so it should be incremented before
+ * interlocks are dropped/preemption is enabled, or
+ * before a CAS is executed to acquire the lock).
+ *
+ */
+
+/*
+ * lck_rw_clear_promotion: Undo priority promotions when the last RW
+ * lock is released by a thread (if a promotion was active)
+ */
+void lck_rw_clear_promotion(thread_t thread)
+{
+       assert(thread->rwlock_count == 0);
+
+       /* Cancel any promotions if the thread had actually blocked while holding a RW lock */
+       spl_t s = splsched();
+
+       thread_lock(thread);
+
+       if (thread->sched_flags & TH_SFLAG_RW_PROMOTED) {
+               thread->sched_flags &= ~TH_SFLAG_RW_PROMOTED;
+
+               if (thread->sched_flags & TH_SFLAG_PROMOTED) {
+                       /* Thread still has a mutex promotion */
+               } else if (thread->sched_flags & TH_SFLAG_DEPRESSED_MASK) {
+                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_RW_DEMOTE) | DBG_FUNC_NONE,
+                                             (uintptr_t)thread_tid(thread), thread->sched_pri, DEPRESSPRI, 0, 0);
+
+                       set_sched_pri(thread, DEPRESSPRI);
+               } else {
+                       KERNEL_DEBUG_CONSTANT(MACHDBG_CODE(DBG_MACH_SCHED, MACH_RW_DEMOTE) | DBG_FUNC_NONE,
+                                             (uintptr_t)thread_tid(thread), thread->sched_pri, thread->base_pri, 0, 0);
+
+                       thread_recompute_sched_pri(thread, FALSE);
+               }
+       }
+
+       thread_unlock(thread);
+       splx(s);
+}
+
+/*
+ * Callout from context switch if the thread goes
+ * off core with a positive rwlock_count
+ *
+ * Called at splsched with the thread locked
+ */
+void
+lck_rw_set_promotion_locked(thread_t thread)
+{
+       if (LcksOpts & disLkRWPrio)
+               return;
+
+       integer_t priority;
+
+       priority = thread->sched_pri;
+
+       if (priority < thread->base_pri)
+               priority = thread->base_pri;
+       if (priority < BASEPRI_BACKGROUND)
+               priority = BASEPRI_BACKGROUND;
+
+       if ((thread->sched_pri < priority) ||
+           !(thread->sched_flags & TH_SFLAG_RW_PROMOTED)) {
+               KERNEL_DEBUG_CONSTANT(
+                       MACHDBG_CODE(DBG_MACH_SCHED, MACH_RW_PROMOTE) | DBG_FUNC_NONE,
+                       (uintptr_t)thread_tid(thread), thread->sched_pri,
+                       thread->base_pri, priority, 0);
+
+               thread->sched_flags |= TH_SFLAG_RW_PROMOTED;
+
+               if (thread->sched_pri < priority)
+                       set_sched_pri(thread, priority);
+       }
+}
+
 kern_return_t
 host_lockgroup_info(
        host_t                                  host,
@@ -794,22 +1457,23 @@ host_lockgroup_info(
        lockgroup_info_t        *lockgroup_info;
        vm_offset_t                     lockgroup_info_addr;
        vm_size_t                       lockgroup_info_size;
+       vm_size_t                       lockgroup_info_vmsize;
        lck_grp_t                       *lck_grp;
        unsigned int            i;
-       vm_size_t                       used;
        vm_map_copy_t           copy;
        kern_return_t           kr;
 
        if (host == HOST_NULL)
                return KERN_INVALID_HOST;
 
-       mutex_lock(&lck_grp_lock);
+       lck_mtx_lock(&lck_grp_lock);
 
-       lockgroup_info_size = round_page(lck_grp_cnt * sizeof *lockgroup_info);
+       lockgroup_info_size = lck_grp_cnt * sizeof(*lockgroup_info);
+       lockgroup_info_vmsize = round_page(lockgroup_info_size);
        kr = kmem_alloc_pageable(ipc_kernel_map,
-                                                &lockgroup_info_addr, lockgroup_info_size);
+                                                &lockgroup_info_addr, lockgroup_info_vmsize, VM_KERN_MEMORY_IPC);
        if (kr != KERN_SUCCESS) {
-               mutex_unlock(&lck_grp_lock);
+               lck_mtx_unlock(&lck_grp_lock);
                return(kr);
        }
 
@@ -853,12 +1517,10 @@ host_lockgroup_info(
        }
 
        *lockgroup_infoCntp = lck_grp_cnt;
-       mutex_unlock(&lck_grp_lock);
+       lck_mtx_unlock(&lck_grp_lock);
 
-       used = (*lockgroup_infoCntp) * sizeof *lockgroup_info;
-
-       if (used != lockgroup_info_size)
-               bzero((char *) lockgroup_info, lockgroup_info_size - used);
+       if (lockgroup_info_size != lockgroup_info_vmsize)
+               bzero((char *)lockgroup_info, lockgroup_info_vmsize - lockgroup_info_size);
 
        kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)lockgroup_info_addr,
                           (vm_map_size_t)lockgroup_info_size, TRUE, &copy);
@@ -870,203 +1532,57 @@ host_lockgroup_info(
 }
 
 /*
- * Compatibility module 
+ * Atomic primitives, prototyped in kern/simple_lock.h
+ * Noret versions are more efficient on some architectures
  */
-
-extern lck_rw_t                *lock_alloc_EXT( boolean_t can_sleep, unsigned short  tag0, unsigned short  tag1);
-extern void            lock_done_EXT(lck_rw_t *lock);
-extern void            lock_free_EXT(lck_rw_t *lock);
-extern void            lock_init_EXT(lck_rw_t *lock, boolean_t can_sleep, unsigned short tag0, unsigned short tag1);
-extern void            lock_read_EXT(lck_rw_t *lock);
-extern boolean_t       lock_read_to_write_EXT(lck_rw_t *lock);
-extern void            lock_write_EXT(lck_rw_t *lock);
-extern void            lock_write_to_read_EXT(lck_rw_t *lock);
-extern wait_result_t   thread_sleep_lock_write_EXT( 
-                               event_t event, lck_rw_t *lock, wait_interrupt_t interruptible);
-
-extern lck_mtx_t       *mutex_alloc_EXT(unsigned short tag);
-extern void            mutex_free_EXT(lck_mtx_t *mutex);
-extern void            mutex_init_EXT(lck_mtx_t *mutex, unsigned short tag);
-extern void            mutex_lock_EXT(lck_mtx_t *mutex);
-extern boolean_t       mutex_try_EXT(lck_mtx_t *mutex);
-extern void            mutex_unlock_EXT(lck_mtx_t *mutex);
-extern wait_result_t   thread_sleep_mutex_EXT(
-                               event_t event, lck_mtx_t *mutex, wait_interrupt_t interruptible);
-extern wait_result_t   thread_sleep_mutex_deadline_EXT(
-                               event_t event, lck_mtx_t *mutex, uint64_t deadline, wait_interrupt_t interruptible);
-
-extern void            usimple_lock_EXT(lck_spin_t *lock);
-extern void            usimple_lock_init_EXT(lck_spin_t *lock, unsigned short tag);
-extern unsigned int    usimple_lock_try_EXT(lck_spin_t *lock);
-extern void            usimple_unlock_EXT(lck_spin_t *lock);
-extern wait_result_t   thread_sleep_usimple_lock_EXT(event_t event, lck_spin_t *lock, wait_interrupt_t interruptible);
-
-lck_rw_t * 
-lock_alloc_EXT(
-       __unused boolean_t       can_sleep,
-       __unused unsigned short  tag0,
-       __unused unsigned short  tag1)
-{
-       return( lck_rw_alloc_init( &LockCompatGroup, LCK_ATTR_NULL));
-}
-
-void
-lock_done_EXT(
-       lck_rw_t        *lock)
-{
-       (void) lck_rw_done(lock);
-}
-
-void
-lock_free_EXT(
-       lck_rw_t        *lock)
-{
-       lck_rw_free(lock, &LockCompatGroup);
-}
-
-void
-lock_init_EXT(
-       lck_rw_t        *lock,
-       __unused boolean_t      can_sleep,
-       __unused unsigned short tag0,
-       __unused unsigned short tag1)
-{
-       lck_rw_init(lock, &LockCompatGroup, LCK_ATTR_NULL);     
-}
-
-void
-lock_read_EXT(
-       lck_rw_t        *lock)
-{
-       lck_rw_lock_shared( lock);
-}
-
-boolean_t
-lock_read_to_write_EXT(
-       lck_rw_t        *lock)
-{
-       return( lck_rw_lock_shared_to_exclusive(lock));
-}
-
-void
-lock_write_EXT(
-       lck_rw_t        *lock)
-{
-       lck_rw_lock_exclusive(lock);
-}
-
-void
-lock_write_to_read_EXT(
-       lck_rw_t        *lock)
+       
+uint32_t
+hw_atomic_add(volatile uint32_t *dest, uint32_t delt)
 {
-       lck_rw_lock_exclusive_to_shared(lock);
+       ALIGN_TEST(dest,uint32_t);
+       return __c11_atomic_fetch_add(ATOMIC_CAST(uint32_t,dest), delt, memory_order_relaxed) + delt;
 }
 
-wait_result_t
-thread_sleep_lock_write_EXT(
-       event_t                 event,
-       lck_rw_t                *lock,
-       wait_interrupt_t        interruptible)
-{
-       return( lck_rw_sleep(lock, LCK_SLEEP_EXCLUSIVE, event, interruptible));
-}
-
-lck_mtx_t *
-mutex_alloc_EXT(
-       __unused unsigned short         tag)
-{
-       return(lck_mtx_alloc_init(&LockCompatGroup, LCK_ATTR_NULL));
-}
-
-void
-mutex_free_EXT(
-       lck_mtx_t               *mutex)
+uint32_t
+hw_atomic_sub(volatile uint32_t *dest, uint32_t delt)
 {
-       lck_mtx_free(mutex, &LockCompatGroup);  
+       ALIGN_TEST(dest,uint32_t);
+       return __c11_atomic_fetch_sub(ATOMIC_CAST(uint32_t,dest), delt, memory_order_relaxed) - delt;
 }
 
-void
-mutex_init_EXT(
-       lck_mtx_t               *mutex,
-       __unused unsigned short tag)
+uint32_t
+hw_atomic_or(volatile uint32_t *dest, uint32_t mask)
 {
-       lck_mtx_init(mutex, &LockCompatGroup, LCK_ATTR_NULL);   
+       ALIGN_TEST(dest,uint32_t);
+       return __c11_atomic_fetch_or(ATOMIC_CAST(uint32_t,dest), mask, memory_order_relaxed) | mask;
 }
 
 void
-mutex_lock_EXT(
-       lck_mtx_t               *mutex)
+hw_atomic_or_noret(volatile uint32_t *dest, uint32_t mask)
 {
-       lck_mtx_lock(mutex);
+       ALIGN_TEST(dest,uint32_t);
+       __c11_atomic_fetch_or(ATOMIC_CAST(uint32_t,dest), mask, memory_order_relaxed);
 }
 
-boolean_t
-mutex_try_EXT(
-       lck_mtx_t               *mutex)
+uint32_t
+hw_atomic_and(volatile uint32_t *dest, uint32_t mask)
 {
-       return(lck_mtx_try_lock(mutex));
+       ALIGN_TEST(dest,uint32_t);
+       return __c11_atomic_fetch_and(ATOMIC_CAST(uint32_t,dest), mask, memory_order_relaxed) & mask;
 }
 
 void
-mutex_unlock_EXT(
-       lck_mtx_t               *mutex)
-{
-       lck_mtx_unlock(mutex);
-}
-
-wait_result_t
-thread_sleep_mutex_EXT(
-       event_t                 event,
-       lck_mtx_t               *mutex,
-       wait_interrupt_t        interruptible)
+hw_atomic_and_noret(volatile uint32_t *dest, uint32_t mask)
 {
-       return( lck_mtx_sleep(mutex, LCK_SLEEP_DEFAULT, event, interruptible));
+       ALIGN_TEST(dest,uint32_t);
+       __c11_atomic_fetch_and(ATOMIC_CAST(uint32_t,dest), mask, memory_order_relaxed);
 }
 
-wait_result_t
-thread_sleep_mutex_deadline_EXT(
-       event_t                 event,
-       lck_mtx_t               *mutex,
-       uint64_t                deadline,
-       wait_interrupt_t        interruptible)
+uint32_t
+hw_compare_and_store(uint32_t oldval, uint32_t newval, volatile uint32_t *dest)
 {
-       return( lck_mtx_sleep_deadline(mutex, LCK_SLEEP_DEFAULT, event, interruptible, deadline));
+       ALIGN_TEST(dest,uint32_t);
+       return __c11_atomic_compare_exchange_strong(ATOMIC_CAST(uint32_t,dest), &oldval, newval,
+                       memory_order_acq_rel_smp, memory_order_relaxed);
 }
 
-void
-usimple_lock_EXT(
-       lck_spin_t              *lock)
-{
-       lck_spin_lock(lock);
-}
-
-void
-usimple_lock_init_EXT(
-       lck_spin_t              *lock,
-       __unused unsigned short tag)
-{
-       lck_spin_init(lock, &LockCompatGroup, LCK_ATTR_NULL);
-}
-
-unsigned int
-usimple_lock_try_EXT(
-       lck_spin_t              *lock)
-{
-       return(lck_spin_try_lock(lock));
-}
-
-void
-usimple_unlock_EXT(
-       lck_spin_t              *lock)
-{
-       lck_spin_unlock(lock);
-}
-
-wait_result_t
-thread_sleep_usimple_lock_EXT(
-       event_t                 event,
-       lck_spin_t              *lock,
-       wait_interrupt_t        interruptible)
-{
-       return( lck_spin_sleep(lock, LCK_SLEEP_DEFAULT, event, interruptible));
-}