]> git.saurik.com Git - apple/libc.git/blobdiff - pthreads/pthread.c
Libc-763.13.tar.gz
[apple/libc.git] / pthreads / pthread.c
index 0b9edad165f8fcf05b8a3c24ba205320f3fe328d..ef4527e349b264de1238f39b55b1f4f2ee449412 100644 (file)
@@ -1,3 +1,25 @@
+/*
+ * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
+ *
+ * @APPLE_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. 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_LICENSE_HEADER_END@
+ */
 /*
  * Copyright 1996 1995 by Open Software Foundation, Inc. 1997 1996 1995 1994 1993 1992 1991  
  *              All Rights Reserved 
  * POSIX Pthread Library
  */
 
-#define __POSIX_LIB__
+#include "pthread_internals.h"
+#include "pthread_workqueue.h"
+
 #include <assert.h>
 #include <stdio.h>     /* For printf(). */
 #include <stdlib.h>
 #include <errno.h>     /* For __mach_errno_addr() prototype. */
+#include <signal.h>
 #include <sys/time.h>
 #include <sys/resource.h>
 #include <sys/sysctl.h>
-#include <sys/syscall.h>
+#include <sys/queue.h>
+#include <sys/mman.h>
 #include <machine/vmparam.h>
 #include <mach/vm_statistics.h>
+#include <mach/mach_init.h>
 #define        __APPLE_API_PRIVATE
 #include <machine/cpu_capabilities.h>
+#include <libkern/OSAtomic.h>
+#if defined(__ppc__)
+#include <libkern/OSCrossEndian.h>
+#endif
 
-#include "pthread_internals.h"
+
+extern int _pthread_setcancelstate_internal(int state, int *oldstate, int conforming);
+extern int __pthread_sigmask(int, const sigset_t *, sigset_t *);
+
+#ifndef BUILDING_VARIANT /* [ */
+
+__private_extern__ struct __pthread_list __pthread_head = TAILQ_HEAD_INITIALIZER(__pthread_head);
+
+
+
+int32_t workq_targetconc[WORKQ_NUM_PRIOQUEUE];
 
 /* Per-thread kernel support */
 extern void _pthread_set_self(pthread_t);
 extern void mig_init(int);
+static int _pthread_create_pthread_onstack(pthread_attr_t *attrs, void **stack, pthread_t *thread);
+static kern_return_t _pthread_free_pthread_onstack(pthread_t t, int freestruct, int termthread);
+static void _pthread_struct_init(pthread_t t, const pthread_attr_t *attrs, void * stack, size_t stacksize, int kernalloc, int nozero);
+static int  _new_pthread_create_suspended(pthread_t *thread, 
+              const pthread_attr_t *attr,
+              void *(*start_routine)(void *), 
+              void *arg,
+               int create_susp);
 
 /* Get CPU capabilities from the kernel */
 __private_extern__ void _init_cpu_capabilities(void);
@@ -55,6 +104,15 @@ extern void set_malloc_singlethreaded(int);
 
 /* Used when we need to call into the kernel with no reply port */
 extern pthread_lock_t reply_port_lock;
+int _pthread_find_thread(pthread_t thread);
+
+/* Mach message used to notify that a thread needs to be reaped */
+
+typedef struct _pthread_reap_msg_t {
+       mach_msg_header_t header;
+       pthread_t thread;
+       mach_msg_trailer_t trailer;
+} pthread_reap_msg_t;
 
 /* We'll implement this when the main thread is a pthread */
 /* Use the local _pthread struct to avoid malloc before our MiG reply port is set */
@@ -64,12 +122,19 @@ static struct _pthread _thread = {0};
 ** pthread has been created.
 */
 int __is_threaded = 0;
+/* _pthread_count is protected by _pthread_list_lock */
 static int _pthread_count = 1;
+int __unix_conforming = 0;
+__private_extern__ size_t pthreadsize = 0;
+
+/* under rosetta we will use old style creation of threads */
+static int __oldstyle = 0;
 
-static pthread_lock_t _pthread_count_lock = LOCK_INITIALIZER;
+__private_extern__ pthread_lock_t _pthread_list_lock = LOCK_INITIALIZER;
 
 /* Same implementation as LOCK, but without the __is_threaded check */
 int _spin_tries = 0;
+extern kern_return_t syscall_thread_switch(mach_port_name_t, int, mach_msg_timeout_t);
 __private_extern__ void _spin_lock_retry(pthread_lock_t *lock)
 { 
        int tries = _spin_tries;
@@ -81,7 +146,7 @@ __private_extern__ void _spin_lock_retry(pthread_lock_t *lock)
        } while(!_spin_lock_try(lock));
 }
 
-extern mach_port_t thread_recycle_port;
+static mach_port_t thread_recycle_port = MACH_PORT_NULL;
 
 /* These are used to keep track of a semaphore pool shared by mutexes and condition
 ** variables.
@@ -97,177 +162,307 @@ static int max_priority;
 static int min_priority;
 static int pthread_concurrency;
 
-/*
- * [Internal] stack support
- */
-size_t _pthread_stack_size = 0;
-#define STACK_LOWEST(sp)       ((sp) & ~__pthread_stack_mask)
-#define STACK_RESERVED         (sizeof (struct _pthread))
-
-#ifdef STACK_GROWS_UP
-
-/* The stack grows towards higher addresses:
-   |struct _pthread|user stack---------------->|
-   ^STACK_BASE     ^STACK_START
-   ^STACK_SELF
-   ^STACK_LOWEST  */
-#define STACK_BASE(sp)         STACK_LOWEST(sp)
-#define STACK_START(stack_low) (STACK_BASE(stack_low) + STACK_RESERVED)
-#define STACK_SELF(sp)         STACK_BASE(sp)
-
-#else
-
-/* The stack grows towards lower addresses:
-   |<----------------user stack|struct _pthread|
-   ^STACK_LOWEST               ^STACK_START    ^STACK_BASE
-                              ^STACK_SELF  */
-
-#define STACK_BASE(sp)         (((sp) | __pthread_stack_mask) + 1)
-#define STACK_START(stack_low) (STACK_BASE(stack_low) - STACK_RESERVED)
-#define STACK_SELF(sp)         STACK_START(sp)
+static OSSpinLock __workqueue_list_lock = OS_SPINLOCK_INIT;
 
-#endif
+static void _pthread_exit(pthread_t self, void *value_ptr) __dead2;
+static void _pthread_setcancelstate_exit(pthread_t self, void  *value_ptr, int conforming);
+static pthread_attr_t _pthread_attr_default = {0};
+static void _pthread_workq_init(pthread_workqueue_t wq, const pthread_workqueue_attr_t * attr);
+static int kernel_workq_setup = 0;
+static volatile int32_t kernel_workq_count = 0;
+static volatile unsigned int user_workq_count = 0;     /* number of outstanding workqueues */
+static volatile unsigned int user_workitem_count = 0;  /* number of outstanding workitems */
+#define KERNEL_WORKQ_ELEM_MAX          64              /* Max number of elements in the kerrel */
+static int wqreadyprio = 0;    /* current highest prio queue ready with items */
+
+__private_extern__ struct __pthread_workitem_pool __pthread_workitem_pool_head = TAILQ_HEAD_INITIALIZER(__pthread_workitem_pool_head);
+__private_extern__ struct __pthread_workqueue_pool __pthread_workqueue_pool_head = TAILQ_HEAD_INITIALIZER(__pthread_workqueue_pool_head);
+
+static struct _pthread_workitem *  __workqueue_pool_ptr;
+static size_t __workqueue_pool_size = 0;
+static int __workqueue_nitems = 0;
+
+struct _pthread_workqueue_head __pthread_workq0_head;
+struct _pthread_workqueue_head __pthread_workq1_head;
+struct _pthread_workqueue_head __pthread_workq2_head;
+struct _pthread_workqueue_head __pthread_workq3_head;
+pthread_workqueue_head_t __pthread_wq_head_tbl[WORKQ_NUM_PRIOQUEUE] = {&__pthread_workq0_head, &__pthread_workq1_head, &__pthread_workq2_head, &__pthread_workq3_head};
+
+static void workqueue_list_lock(void);
+static void workqueue_list_unlock(void);
+static int valid_workq(pthread_workqueue_t);
+static void pick_nextworkqueue_droplock(void);
+static int post_nextworkitem(pthread_workqueue_t workq);
+static void _pthread_workq_return(pthread_t self);
+static pthread_workqueue_attr_t _pthread_wq_attr_default = {0};
+extern void start_wqthread(pthread_t self, mach_port_t kport, void * stackaddr, pthread_workitem_t item, int reuse);
+extern void thread_start(pthread_t self, mach_port_t kport, void *(*fun)(void *), void * funarg, size_t stacksize, unsigned int flags);
+static pthread_workitem_t alloc_workitem(void);
+static void free_workitem(pthread_workitem_t);
+static void grow_workitem(void);
+static pthread_workqueue_t alloc_workqueue(void);
+static void free_workqueue(pthread_workqueue_t);
+static int _pthread_work_internal_init(void);
+static void workqueue_exit(pthread_t self, pthread_workqueue_t workq, pthread_workitem_t item);
+void _pthread_fork_child_postinit();
+
+void pthread_workqueue_atfork_prepare(void);
+void pthread_workqueue_atfork_parent(void);
+void pthread_workqueue_atfork_child(void);
+
+extern void dispatch_atfork_prepare(void);
+extern void dispatch_atfork_parent(void);
+extern void dispatch_atfork_child(void);
+
+/* workq_kernreturn commands */
+#define WQOPS_QUEUE_ADD 1
+#define WQOPS_QUEUE_REMOVE 2
+#define WQOPS_THREAD_RETURN 4
+#define WQOPS_THREAD_SETCONC  8
 
-#if defined(__ppc__)
+/*
+ * Flags filed passed to bsdthread_create and back in pthread_start 
+31  <---------------------------------> 0
+_________________________________________
+| flags(8) | policy(8) | importance(16) |
+-----------------------------------------
+*/
+__private_extern__
+void _pthread_start(pthread_t self, mach_port_t kport, void *(*fun)(void *), void * funarg, size_t stacksize, unsigned int flags);
+
+__private_extern__
+void _pthread_wqthread(pthread_t self, mach_port_t kport, void * stackaddr, pthread_workitem_t item, int reuse);
+
+#define PTHREAD_START_CUSTOM   0x01000000
+#define PTHREAD_START_SETSCHED 0x02000000
+#define PTHREAD_START_DETACHED 0x04000000
+#define PTHREAD_START_POLICY_BITSHIFT 16
+#define PTHREAD_START_POLICY_MASK 0xff
+#define PTHREAD_START_IMPORTANCE_MASK 0xffff
+
+static int pthread_setschedparam_internal(pthread_t, mach_port_t, int, const struct sched_param *);
+extern pthread_t __bsdthread_create(void *(*func)(void *), void * func_arg, void * stack, pthread_t  thread, unsigned int flags);
+extern int __bsdthread_register(void (*)(pthread_t, mach_port_t, void *(*)(void *), void *, size_t, unsigned int), void (*)(pthread_t, mach_port_t, void *, pthread_workitem_t, int), int,void (*)(pthread_t, mach_port_t, void *(*)(void *), void *, size_t, unsigned int), int32_t *,__uint64_t);
+extern int __bsdthread_terminate(void * freeaddr, size_t freesize, mach_port_t kport, mach_port_t joinsem);
+extern __uint64_t __thread_selfid( void );
+extern int __pthread_canceled(int);
+extern void _pthread_keys_init(void);
+extern int __pthread_kill(mach_port_t, int);
+extern int __pthread_markcancel(int);
+extern int __workq_open(void);
+
+#define WORKQUEUE_OVERCOMMIT 0x10000
+
+extern int __workq_kernreturn(int, pthread_workitem_t, int, int);
+
+#if defined(__ppc__) || defined(__ppc64__)
 static const vm_address_t PTHREAD_STACK_HINT = 0xF0000000;
-#elif defined(__i386__)
+#elif defined(__i386__) || defined(__x86_64__)
 static const vm_address_t PTHREAD_STACK_HINT = 0xB0000000;
+#elif defined(__arm__)
+static const vm_address_t PTHREAD_STACK_HINT = 0x30000000;
 #else
 #error Need to define a stack address hint for this architecture
 #endif
 
-/* Set the base address to use as the stack pointer, before adjusting due to the ABI */
+/* Set the base address to use as the stack pointer, before adjusting due to the ABI
+ * The guardpages for stackoverflow protection is also allocated here
+ * If the stack was already allocated(stackaddr in attr)  then there are no guardpages 
+ * set up for the thread
+ */
 
 static int
 _pthread_allocate_stack(pthread_attr_t *attrs, void **stack)
 {
     kern_return_t kr;
-#if 1
+    vm_address_t stackaddr;
+    size_t guardsize;
+
     assert(attrs->stacksize >= PTHREAD_STACK_MIN);
     if (attrs->stackaddr != NULL) {
-        assert(((vm_address_t)(attrs->stackaddr) & (vm_page_size - 1)) == 0);
+       /* No guard pages setup in this case */
+        assert(((uintptr_t)attrs->stackaddr % vm_page_size) == 0);
         *stack = attrs->stackaddr;
          return 0;
     }
 
-    *((vm_address_t *)stack) = PTHREAD_STACK_HINT;
-    kr = vm_map(mach_task_self(), (vm_address_t *)stack,
-                       attrs->stacksize + vm_page_size,
+    guardsize = attrs->guardsize;
+    stackaddr = PTHREAD_STACK_HINT;
+    kr = vm_map(mach_task_self(), &stackaddr,
+                       attrs->stacksize + guardsize,
                        vm_page_size-1,
                        VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE , MEMORY_OBJECT_NULL,
                        0, FALSE, VM_PROT_DEFAULT, VM_PROT_ALL,
                        VM_INHERIT_DEFAULT);
     if (kr != KERN_SUCCESS)
        kr = vm_allocate(mach_task_self(),
-                                       (vm_address_t *)stack, attrs->stacksize + vm_page_size,
+                                       &stackaddr, attrs->stacksize + guardsize,
                                        VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE);
     if (kr != KERN_SUCCESS) {
         return EAGAIN;
     }
- #ifdef STACK_GROWS_UP
-     /* The guard page is the page one higher than the stack */
-     /* The stack base is at the lowest address */
-    kr = vm_protect(mach_task_self(), *stack + attrs->stacksize, vm_page_size, FALSE, VM_PROT_NONE);
- #else
      /* The guard page is at the lowest address */
      /* The stack base is the highest address */
-    kr = vm_protect(mach_task_self(), (vm_address_t)*stack, vm_page_size, FALSE, VM_PROT_NONE);
-    *stack += attrs->stacksize + vm_page_size;
- #endif
+    if (guardsize)
+       kr = vm_protect(mach_task_self(), stackaddr, guardsize, FALSE, VM_PROT_NONE);
+    *stack = (void *)(stackaddr + attrs->stacksize + guardsize);
+        return 0;
+}
 
-#else
-    vm_address_t cur_stack = (vm_address_t)0;
-       if (free_stacks == 0)
-       {
-           /* Allocating guard pages is done by doubling
-            * the actual stack size, since STACK_BASE() needs
-            * to have stacks aligned on stack_size. Allocating just 
-            * one page takes as much memory as allocating more pages
-            * since it will remain one entry in the vm map.
-            * Besides, allocating more than one page allows tracking the
-            * overflow pattern when the overflow is bigger than one page.
-            */
-#ifndef        NO_GUARD_PAGES
-# define       GUARD_SIZE(a)   (2*(a))
-# define       GUARD_MASK(a)   (((a)<<1) | 1)
-#else
-# define       GUARD_SIZE(a)   (a)
-# define       GUARD_MASK(a)   (a)
+static int
+_pthread_create_pthread_onstack(pthread_attr_t *attrs, void **stack, pthread_t *thread)
+{
+    kern_return_t kr;
+       pthread_t t;
+    vm_address_t stackaddr;
+    size_t guardsize, allocsize;
+
+       assert(attrs->stacksize >= PTHREAD_STACK_MIN);
+
+    if (attrs->stackaddr != NULL) {
+       /* No guard pages setup in this case */
+               assert(((uintptr_t)attrs->stackaddr % vm_page_size) == 0);
+               *stack = attrs->stackaddr;
+               t = (pthread_t)malloc(pthreadsize);
+               _pthread_struct_init(t, attrs, attrs->stackaddr, 0, 0, 0);
+               t->freeStackOnExit = 0;
+               t->freeaddr = 0;
+               t->freesize = 0;
+               *thread = t;
+               return 0;
+       }
+
+    guardsize = attrs->guardsize;
+       allocsize = attrs->stacksize + guardsize + pthreadsize;
+    stackaddr = PTHREAD_STACK_HINT;
+    kr = vm_map(mach_task_self(), &stackaddr,
+                       allocsize,
+                       vm_page_size-1,
+                       VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE , MEMORY_OBJECT_NULL,
+                       0, FALSE, VM_PROT_DEFAULT, VM_PROT_ALL,
+                       VM_INHERIT_DEFAULT);
+    if (kr != KERN_SUCCESS)
+       kr = vm_allocate(mach_task_self(),
+                                       &stackaddr, allocsize,
+                                       VM_MAKE_TAG(VM_MEMORY_STACK)| VM_FLAGS_ANYWHERE);
+    if (kr != KERN_SUCCESS) {
+        return EAGAIN;
+    }
+     /* The guard page is at the lowest address */
+     /* The stack base is the highest address */
+    if (guardsize)
+       kr = vm_protect(mach_task_self(), stackaddr, guardsize, FALSE, VM_PROT_NONE);
+
+
+       *stack = (void *)(stackaddr + attrs->stacksize + guardsize);
+
+       t = (pthread_t)(stackaddr + attrs->stacksize + guardsize);
+       _pthread_struct_init(t, attrs, *stack, 0, 0, 1);
+       t->kernalloc = 0;
+       t->freesize = allocsize;
+       t->freeaddr = (void *)stackaddr;
+       t->freeStackOnExit = 1;
+       *thread = t;
+
+    return 0;
+}
+
+static kern_return_t
+_pthread_free_pthread_onstack(pthread_t t, int freestruct, int termthread)
+{
+       kern_return_t res = 0;
+       vm_address_t freeaddr;
+       size_t freesize;
+       int thread_count;
+       mach_port_t kport;
+       semaphore_t joinsem = SEMAPHORE_NULL;
+
+#if PTH_TRACE
+       __kdebug_trace(0x900001c, freestruct, termthread, 0, 0, 0);
 #endif
-               while (lowest_stack > GUARD_SIZE(__pthread_stack_size))
-               {
-                       lowest_stack -= GUARD_SIZE(__pthread_stack_size);
-                       /* Ensure stack is there */
-                       kr = vm_allocate(mach_task_self(),
-                                        &lowest_stack,
-                                        GUARD_SIZE(__pthread_stack_size),
-                                        FALSE);
-#ifndef        NO_GUARD_PAGES
-                       if (kr == KERN_SUCCESS) {
-# ifdef        STACK_GROWS_UP
-                           kr = vm_protect(mach_task_self(),
-                                           lowest_stack+__pthread_stack_size,
-                                           __pthread_stack_size,
-                                           FALSE, VM_PROT_NONE);
-# else /* STACK_GROWS_UP */
-                           kr = vm_protect(mach_task_self(),
-                                           lowest_stack,
-                                           __pthread_stack_size,
-                                           FALSE, VM_PROT_NONE);
-                           lowest_stack += __pthread_stack_size;
-# endif        /* STACK_GROWS_UP */
-                           if (kr == KERN_SUCCESS)
-                               break;
+       kport = t->kernel_thread;
+       joinsem = t->joiner_notify;
+
+       if (t->freeStackOnExit) {
+               freeaddr = (vm_address_t)t->freeaddr;
+               if (freestruct)
+                       freesize = t->stacksize + t->guardsize + pthreadsize;
+               else
+                       freesize = t->stacksize + t->guardsize;
+               if (termthread) {
+                       mig_dealloc_reply_port(MACH_PORT_NULL);
+                       LOCK(_pthread_list_lock);
+                       if (freestruct != 0) {
+                               TAILQ_REMOVE(&__pthread_head, t, plist);
+                               /* if parent has not returned from create yet keep pthread_t */
+#if PTH_LISTTRACE
+                               __kdebug_trace(0x9000010, t, 0, 0, 1, 0);
+#endif
+                               if (t->parentcheck == 0)
+                                       freesize -= pthreadsize;
                        }
-#else
-                       if (kr == KERN_SUCCESS)
-                           break;
+                       t->childexit = 1;
+                       thread_count = --_pthread_count;
+                       UNLOCK(_pthread_list_lock);
+
+#if PTH_TRACE
+               __kdebug_trace(0x9000020, freeaddr, freesize, kport, 1, 0);
 #endif
+                       if (thread_count <=0)
+                               exit(0);
+                       else
+                               __bsdthread_terminate((void *)freeaddr, freesize, kport, joinsem);
+                       LIBC_ABORT("thread %p didn't terminate", t);
+               } else  {
+#if PTH_TRACE
+                       __kdebug_trace(0x9000024, freeaddr, freesize, 0, 1, 0);
+#endif
+                       res = vm_deallocate(mach_task_self(), freeaddr, freesize);
                }
-               if (lowest_stack > 0)
-                       free_stacks = (vm_address_t *)lowest_stack;
-               else
-               {
-                       /* Too bad.  We'll just have to take what comes.
-                          Use vm_map instead of vm_allocate so we can
-                          specify alignment.  */
-                       kr = vm_map(mach_task_self(), &lowest_stack,
-                                   GUARD_SIZE(__pthread_stack_size),
-                                   GUARD_MASK(__pthread_stack_mask),
-                                   TRUE /* anywhere */, MEMORY_OBJECT_NULL,
-                                   0, FALSE, VM_PROT_DEFAULT, VM_PROT_ALL,
-                                   VM_INHERIT_DEFAULT);
-                       /* This really shouldn't fail and if it does I don't
-                          know what to do.  */
-#ifndef        NO_GUARD_PAGES
-                       if (kr == KERN_SUCCESS) {
-# ifdef        STACK_GROWS_UP
-                           kr = vm_protect(mach_task_self(),
-                                           lowest_stack+__pthread_stack_size,
-                                           __pthread_stack_size,
-                                           FALSE, VM_PROT_NONE);
-# else /* STACK_GROWS_UP */
-                           kr = vm_protect(mach_task_self(),
-                                           lowest_stack,
-                                           __pthread_stack_size,
-                                           FALSE, VM_PROT_NONE);
-                           lowest_stack += __pthread_stack_size;
-# endif        /* STACK_GROWS_UP */
+       } else {
+               if (termthread) {
+                       mig_dealloc_reply_port(MACH_PORT_NULL);
+                       LOCK(_pthread_list_lock);
+                       if (freestruct != 0) {
+                               TAILQ_REMOVE(&__pthread_head, t, plist);
+#if PTH_LISTTRACE
+                               __kdebug_trace(0x9000010, t, 0, 0, 2, 0);
+#endif
+                       }
+                       thread_count = --_pthread_count;
+                       t->childexit = 1;
+                       UNLOCK(_pthread_list_lock);
+
+                       if (freestruct)  {
+#if PTH_TRACE
+               __kdebug_trace(0x9000008, t, 0, 0, 2, 0);
+#endif
+                               free(t);
                        }
+
+                       freeaddr = 0;
+                       freesize = 0;
+#if PTH_TRACE
+               __kdebug_trace(0x9000020, 0, 0, kport, 2, 0);
+#endif
+
+                       if (thread_count <=0)
+                               exit(0);
+                       else
+                               __bsdthread_terminate(NULL, 0, kport, joinsem);
+                       LIBC_ABORT("thread %p didn't terminate", t);
+               } else if (freestruct) {
+                      t->sig = _PTHREAD_NO_SIG;
+#if PTH_TRACE
+                       __kdebug_trace(0x9000024, t, 0, 0, 2, 0);
 #endif
-                       free_stacks = (vm_address_t *)lowest_stack;
-                       lowest_stack = 0;
+                       free(t);
                }
-               *free_stacks = 0; /* No other free stacks */
        }
-       cur_stack = STACK_START((vm_address_t) free_stacks);
-       free_stacks = (vm_address_t *)*free_stacks;
-       cur_stack = _adjust_sp(cur_stack); /* Machine dependent stack fudging */
-#endif
-        return 0;
-}
+       return(res);
+} 
+
 
-static pthread_attr_t _pthread_attr_default = {0};
 
 /*
  * Destroy a thread attribute structure
@@ -277,7 +472,8 @@ pthread_attr_destroy(pthread_attr_t *attr)
 {
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
-               return (ESUCCESS);
+               attr->sig = 0;
+               return (0);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
@@ -295,7 +491,7 @@ pthread_attr_getdetachstate(const pthread_attr_t *attr,
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *detachstate = attr->detached;
-               return (ESUCCESS);
+               return (0);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
@@ -313,7 +509,7 @@ pthread_attr_getinheritsched(const pthread_attr_t *attr,
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *inheritsched = attr->inherit;
-               return (ESUCCESS);
+               return (0);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
@@ -331,7 +527,7 @@ pthread_attr_getschedparam(const pthread_attr_t *attr,
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *param = attr->param;
-               return (ESUCCESS);
+               return (0);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
@@ -349,30 +545,34 @@ pthread_attr_getschedpolicy(const pthread_attr_t *attr,
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *policy = attr->policy;
-               return (ESUCCESS);
+               return (0);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
 }
 
-static const size_t DEFAULT_STACK_SIZE = DFLSSIZ;
+/* Retain the existing stack size of 512K and not depend on Main thread default stack size */
+static const size_t DEFAULT_STACK_SIZE = (512*1024);
 /*
  * Initialize a thread attribute structure to default values.
  */
 int       
 pthread_attr_init(pthread_attr_t *attr)
 {
-        attr->stacksize = DEFAULT_STACK_SIZE;
-        attr->stackaddr = NULL;
+       attr->stacksize = DEFAULT_STACK_SIZE;
+       attr->stackaddr = NULL;
        attr->sig = _PTHREAD_ATTR_SIG;
-       attr->policy = _PTHREAD_DEFAULT_POLICY;
        attr->param.sched_priority = default_priority;
        attr->param.quantum = 10; /* quantum isn't public yet */
-       attr->inherit = _PTHREAD_DEFAULT_INHERITSCHED;
        attr->detached = PTHREAD_CREATE_JOINABLE;
-        attr->freeStackOnExit = TRUE;
-       return (ESUCCESS);
+       attr->inherit = _PTHREAD_DEFAULT_INHERITSCHED;
+       attr->policy = _PTHREAD_DEFAULT_POLICY;
+       attr->freeStackOnExit = 1;
+       attr->fastpath = 1;
+       attr->schedset = 0;
+       attr->guardsize = vm_page_size;
+       return (0);
 }
 
 /*
@@ -389,7 +589,7 @@ pthread_attr_setdetachstate(pthread_attr_t *attr,
                    (detachstate == PTHREAD_CREATE_DETACHED))
                {
                        attr->detached = detachstate;
-                       return (ESUCCESS);
+                       return (0);
                } else
                {
                        return (EINVAL);
@@ -414,7 +614,7 @@ pthread_attr_setinheritsched(pthread_attr_t *attr,
                    (inheritsched == PTHREAD_EXPLICIT_SCHED))
                {
                        attr->inherit = inheritsched;
-                       return (ESUCCESS);
+                       return (0);
                } else
                {
                        return (EINVAL);
@@ -437,7 +637,8 @@ pthread_attr_setschedparam(pthread_attr_t *attr,
        {
                /* TODO: Validate sched_param fields */
                attr->param = *param;
-               return (ESUCCESS);
+               attr->schedset = 1;
+               return (0);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
@@ -459,7 +660,8 @@ pthread_attr_setschedpolicy(pthread_attr_t *attr,
                    (policy == SCHED_FIFO))
                {
                        attr->policy = policy;
-                       return (ESUCCESS);
+                       attr->schedset = 1;
+                       return (0);
                } else
                {
                        return (EINVAL);
@@ -481,7 +683,7 @@ pthread_attr_setscope(pthread_attr_t *attr,
     if (attr->sig == _PTHREAD_ATTR_SIG) {
         if (scope == PTHREAD_SCOPE_SYSTEM) {
             /* No attribute yet for the scope */
-            return (ESUCCESS);
+            return (0);
         } else if (scope == PTHREAD_SCOPE_PROCESS) {
             return (ENOTSUP);
         }
@@ -494,12 +696,12 @@ pthread_attr_setscope(pthread_attr_t *attr,
  * We currently only provide PTHREAD_SCOPE_SYSTEM
  */
 int
-pthread_attr_getscope(pthread_attr_t *attr,
+pthread_attr_getscope(const pthread_attr_t *attr,
                             int *scope)
 {
     if (attr->sig == _PTHREAD_ATTR_SIG) {
         *scope = PTHREAD_SCOPE_SYSTEM;
-        return (ESUCCESS);
+        return (0);
     }
     return (EINVAL); /* Not an attribute structure! */
 }
@@ -510,7 +712,7 @@ pthread_attr_getstackaddr(const pthread_attr_t *attr, void **stackaddr)
 {
     if (attr->sig == _PTHREAD_ATTR_SIG) {
         *stackaddr = attr->stackaddr;
-        return (ESUCCESS);
+        return (0);
     } else {
         return (EINVAL); /* Not an attribute structure! */
     }
@@ -519,10 +721,11 @@ pthread_attr_getstackaddr(const pthread_attr_t *attr, void **stackaddr)
 int
 pthread_attr_setstackaddr(pthread_attr_t *attr, void *stackaddr)
 {
-    if ((attr->sig == _PTHREAD_ATTR_SIG) && (((vm_offset_t)stackaddr & (vm_page_size - 1)) == 0)) {
+    if ((attr->sig == _PTHREAD_ATTR_SIG) && (((uintptr_t)stackaddr % vm_page_size) == 0)) {
         attr->stackaddr = stackaddr;
-        attr->freeStackOnExit = FALSE;
-        return (ESUCCESS);
+        attr->freeStackOnExit = 0;
+       attr->fastpath = 0;
+        return (0);
     } else {
         return (EINVAL); /* Not an attribute structure! */
     }
@@ -533,7 +736,7 @@ pthread_attr_getstacksize(const pthread_attr_t *attr, size_t *stacksize)
 {
     if (attr->sig == _PTHREAD_ATTR_SIG) {
         *stacksize = attr->stacksize;
-        return (ESUCCESS);
+        return (0);
     } else {
         return (EINVAL); /* Not an attribute structure! */
     }
@@ -544,7 +747,7 @@ pthread_attr_setstacksize(pthread_attr_t *attr, size_t stacksize)
 {
     if ((attr->sig == _PTHREAD_ATTR_SIG) && ((stacksize % vm_page_size) == 0) && (stacksize >= PTHREAD_STACK_MIN)) {
         attr->stacksize = stacksize;
-        return (ESUCCESS);
+        return (0);
     } else {
         return (EINVAL); /* Not an attribute structure! */
     }
@@ -554,29 +757,68 @@ int
 pthread_attr_getstack(const pthread_attr_t *attr, void **stackaddr, size_t * stacksize)
 {
     if (attr->sig == _PTHREAD_ATTR_SIG) {
-        *stackaddr = attr->stackaddr;
+       *stackaddr = (void *)((uintptr_t)attr->stackaddr - attr->stacksize);
         *stacksize = attr->stacksize;
-        return (ESUCCESS);
+        return (0);
     } else {
         return (EINVAL); /* Not an attribute structure! */
     }
 }
 
+/* By SUSV spec, the stackaddr is the base address, the lowest addressable 
+ * byte address. This is not the same as in pthread_attr_setstackaddr.
+ */
 int
 pthread_attr_setstack(pthread_attr_t *attr, void *stackaddr, size_t stacksize)
 {
     if ((attr->sig == _PTHREAD_ATTR_SIG) && 
-       (((vm_offset_t)stackaddr & (vm_page_size - 1)) == 0) && 
-       ((stacksize % vm_page_size) == 0) && (stacksize >= PTHREAD_STACK_MIN)) {
-               attr->stackaddr = stackaddr;
-               attr->freeStackOnExit = FALSE;
+       (((uintptr_t)stackaddr % vm_page_size) == 0) && 
+        ((stacksize % vm_page_size) == 0) && (stacksize >= PTHREAD_STACK_MIN)) {
+               attr->stackaddr = (void *)((uintptr_t)stackaddr + stacksize);
                attr->stacksize = stacksize;
-               return (ESUCCESS);
+               attr->freeStackOnExit = 0;
+               attr->fastpath = 0;
+               return (0);
     } else {
         return (EINVAL); /* Not an attribute structure! */
     }
 }
 
+
+/*
+ * Set the guardsize attribute in the attr.
+ */
+int
+pthread_attr_setguardsize(pthread_attr_t *attr,
+                            size_t guardsize)
+{
+    if (attr->sig == _PTHREAD_ATTR_SIG) {
+       /* Guardsize of 0 is valid, ot means no guard */
+        if ((guardsize % vm_page_size) == 0) {
+               attr->guardsize = guardsize;
+               attr->fastpath = 0;
+            return (0);
+       } else
+               return(EINVAL);
+    }
+    return (EINVAL); /* Not an attribute structure! */
+}
+
+/*
+ * Get the guardsize attribute in the attr.
+ */
+int
+pthread_attr_getguardsize(const pthread_attr_t *attr,
+                            size_t *guardsize)
+{
+    if (attr->sig == _PTHREAD_ATTR_SIG) {
+        *guardsize = attr->guardsize;
+        return (0);
+    }
+    return (EINVAL); /* Not an attribute structure! */
+}
+
+
 /*
  * Create and start execution of a new thread.
  */
@@ -585,7 +827,74 @@ static void
 _pthread_body(pthread_t self)
 {
     _pthread_set_self(self);
-    pthread_exit((self->fun)(self->arg));
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+       if( (self->thread_id = __thread_selfid()) == (__uint64_t)-1)
+               printf("Failed to set thread_id in _pthread_body\n");
+#endif
+    _pthread_exit(self, (self->fun)(self->arg));
+}
+
+void
+_pthread_start(pthread_t self, mach_port_t kport, void *(*fun)(void *), void * funarg, size_t stacksize, unsigned int pflags)
+{
+#if WQ_DEBUG
+       pthread_t pself;
+#endif
+       pthread_attr_t *attrs = &_pthread_attr_default;
+       char * stackaddr;
+
+       if ((pflags & PTHREAD_START_CUSTOM) == 0) {
+               stackaddr = (char *)self;
+               _pthread_struct_init(self, attrs, stackaddr,  stacksize, 1, 1);
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+               _pthread_set_self(self);
+#endif
+               LOCK(_pthread_list_lock);
+               if (pflags & PTHREAD_START_SETSCHED) {
+                       self->policy = ((pflags >> PTHREAD_START_POLICY_BITSHIFT) & PTHREAD_START_POLICY_MASK);
+                       self->param.sched_priority = (pflags & PTHREAD_START_IMPORTANCE_MASK);
+               }
+               /* These are not joinable threads */
+               if ((pflags & PTHREAD_START_DETACHED) == PTHREAD_START_DETACHED)  {
+                       self->detached &= ~PTHREAD_CREATE_JOINABLE;
+                       self->detached |= PTHREAD_CREATE_DETACHED;
+               }
+       }  else { 
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+               _pthread_set_self(self);
+#endif
+               LOCK(_pthread_list_lock);
+       }
+       self->kernel_thread = kport;
+       self->fun = fun;
+       self->arg = funarg;
+       
+       /* Add to the pthread list */
+       if (self->parentcheck == 0) {
+               TAILQ_INSERT_TAIL(&__pthread_head, self, plist);
+#if PTH_LISTTRACE
+               __kdebug_trace(0x900000c, self, 0, 0, 3, 0);
+#endif
+               _pthread_count++;
+       }
+       self->childrun = 1;
+       UNLOCK(_pthread_list_lock);
+
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+       if( (self->thread_id = __thread_selfid()) == (__uint64_t)-1)
+               printf("Failed to set thread_id in pthread_start\n");
+#endif
+
+#if WQ_DEBUG
+       pself = pthread_self();
+       if (self != pself)
+               LIBC_ABORT("self %p != pself %p", self, pself);
+#endif
+#if PTH_TRACE
+       __kdebug_trace(0x9000030, self, pflags, 0, 0, 0);
+#endif
+
+       _pthread_exit(self, (self->fun)(self->arg));
 }
 
 int
@@ -595,34 +904,91 @@ _pthread_create(pthread_t t,
                const mach_port_t kernel_thread)
 {
        int res;
-       res = ESUCCESS;
+       res = 0;
        
        do
        {
                memset(t, 0, sizeof(*t));
+               t->newstyle = 0;
+               t->schedset = 0;
+               t->kernalloc = 0;
+               t->tsd[0] = t;
+               t->max_tsd_key = 0;
+               t->wqthread = 0;
+               t->cur_workq = 0;
+               t->cur_workitem = 0;
                t->stacksize = attrs->stacksize;
                 t->stackaddr = (void *)stack;
+               t->guardsize = attrs->guardsize;
                 t->kernel_thread = kernel_thread;
                t->detached = attrs->detached;
                t->inherit = attrs->inherit;
                t->policy = attrs->policy;
                t->param = attrs->param;
-                t->freeStackOnExit = attrs->freeStackOnExit;
-               t->mutexes = (struct _pthread_mutex *)NULL;
+               t->freeStackOnExit = attrs->freeStackOnExit;
+               t->cancel_error = 0;
                t->sig = _PTHREAD_SIG;
                 t->reply_port = MACH_PORT_NULL;
                 t->cthread_self = NULL;
                LOCK_INIT(t->lock);
+               t->plist.tqe_next = (struct _pthread *)0;
+               t->plist.tqe_prev = (struct _pthread **)0;
                t->cancel_state = PTHREAD_CANCEL_ENABLE | PTHREAD_CANCEL_DEFERRED;
-               t->cleanup_stack = (struct _pthread_handler_rec *)NULL;
+               t->__cleanup_stack = (struct __darwin_pthread_handler_rec *)NULL;
                t->death = SEMAPHORE_NULL;
 
                if (kernel_thread != MACH_PORT_NULL)
-                       pthread_setschedparam(t, t->policy, &t->param);
+                       (void)pthread_setschedparam_internal(t, kernel_thread, t->policy, &t->param);
        } while (0);
        return (res);
 }
 
+void
+_pthread_struct_init(pthread_t t, const pthread_attr_t *attrs, void * stack, size_t stacksize, int kernalloc, int nozero)
+{
+       mach_vm_offset_t stackaddr = (mach_vm_offset_t)(uintptr_t)stack;
+
+               if (nozero == 0) {
+                       memset(t, 0, sizeof(*t));
+                       t->plist.tqe_next = (struct _pthread *)0;
+                       t->plist.tqe_prev = (struct _pthread **)0;
+               }
+               t->schedset = attrs->schedset;
+               t->tsd[0] = t;
+               if (kernalloc != 0) {
+                       stackaddr = (mach_vm_offset_t)(uintptr_t)t;
+
+                       /* if allocated from kernel set values appropriately */
+                       t->stacksize = stacksize;
+                               t->stackaddr = (void *)(uintptr_t)stackaddr;
+                       t->freeStackOnExit = 1;
+                       t->freeaddr = (void *)(uintptr_t)(stackaddr - stacksize - vm_page_size);
+                       t->freesize = pthreadsize + stacksize + vm_page_size;
+               } else {
+                       t->stacksize = attrs->stacksize;
+                       t->stackaddr = (void *)stack;
+               }
+               t->guardsize = attrs->guardsize;
+               t->detached = attrs->detached;
+               t->inherit = attrs->inherit;
+               t->policy = attrs->policy;
+               t->param = attrs->param;
+               t->cancel_error = 0;
+               t->sig = _PTHREAD_SIG;
+               t->reply_port = MACH_PORT_NULL;
+               t->cthread_self = NULL;
+               LOCK_INIT(t->lock);
+               t->cancel_state = PTHREAD_CANCEL_ENABLE | PTHREAD_CANCEL_DEFERRED;
+               t->__cleanup_stack = (struct __darwin_pthread_handler_rec *)NULL;
+               t->death = SEMAPHORE_NULL;
+               t->newstyle = 1;
+               t->kernalloc = kernalloc;
+               t->wqthread = 0;
+               t->cur_workq = 0;
+               t->cur_workitem = 0;
+               t->max_tsd_key = 0;
+}
+
 /* Need to deprecate this in future */
 int
 _pthread_is_threaded(void)
@@ -644,25 +1010,94 @@ pthread_is_threaded_np(void)
 mach_port_t
 pthread_mach_thread_np(pthread_t t)
 {
-    thread_t kernel_thread;
+       mach_port_t kport = MACH_PORT_NULL;
+
+       if (t == NULL)
+               goto out;
+       
+       /* 
+        * If the call is on self, return the kernel port. We cannot
+        * add this bypass for main thread as it might have exited,
+        * and we should not return stale port info.
+        */
+       if (t == pthread_self())
+       {
+               kport = t->kernel_thread;
+               goto out;
+       }
+       
+       if (_pthread_lookup_thread(t, &kport, 0) != 0)
+               return((mach_port_t)0);
+
+out:
+       return(kport);
+}
 
-    /* Wait for the creator to initialize it */
-    while ((kernel_thread = t->kernel_thread) == MACH_PORT_NULL)
-       sched_yield();
+pthread_t pthread_from_mach_thread_np(mach_port_t kernel_thread) 
+{
+    struct _pthread * p = NULL;
 
-    return kernel_thread;
+       /* No need to wait as mach port is already known */
+    LOCK(_pthread_list_lock);
+    TAILQ_FOREACH(p, &__pthread_head, plist) {
+        if (p->kernel_thread == kernel_thread) 
+                       break;
+    }
+    UNLOCK(_pthread_list_lock);     
+    return p;
 }
 
 size_t
 pthread_get_stacksize_np(pthread_t t)
 {
-    return t->stacksize;
+       int ret;
+       size_t size = 0;
+
+       if (t == NULL)
+               return(ESRCH);
+       
+       if ( t == pthread_self() || t == &_thread ) //since the main thread will not get de-allocated from underneath us
+       {
+               size=t->stacksize;
+               return size;
+       }
+
+
+       LOCK(_pthread_list_lock);
+
+       if ((ret = _pthread_find_thread(t)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               return(ret);
+       }
+
+       size=t->stacksize;
+       UNLOCK(_pthread_list_lock);
+       
+       return(size);
 }
 
 void *
 pthread_get_stackaddr_np(pthread_t t)
 {
-    return t->stackaddr;
+       int ret;
+       void * addr = NULL;
+
+       if (t == NULL)
+               return((void *)(uintptr_t)ESRCH);
+       
+       if(t == pthread_self() || t == &_thread) //since the main thread will not get deallocated from underneath us
+               return t->stackaddr;
+
+       LOCK(_pthread_list_lock);
+
+       if ((ret = _pthread_find_thread(t)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               return((void *)(uintptr_t)ret);
+       }
+       addr = t->stackaddr;
+       UNLOCK(_pthread_list_lock);
+
+       return(addr);
 }
 
 mach_port_t
@@ -681,20 +1116,93 @@ pthread_main_np(void)
        return ((self->detached & _PTHREAD_CREATE_PARENT) == _PTHREAD_CREATE_PARENT);
 }
 
+
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+/* if we are passed in a pthread_t that is NULL, then we return
+   the current thread's thread_id. So folks don't have to call
+   pthread_self, in addition to us doing it, if they just want 
+   their thread_id.
+*/
+int
+pthread_threadid_np(pthread_t thread, __uint64_t *thread_id)
+{
+       int rval=0;
+       pthread_t self = pthread_self();
+
+       if (thread_id == NULL) {
+               return(EINVAL);
+       } else if (thread == NULL || thread == self) {
+               *thread_id = self->thread_id;
+               return rval;
+       }
+       
+       LOCK(_pthread_list_lock);
+       if ((rval = _pthread_find_thread(thread)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               return(rval);
+       }
+       *thread_id = thread->thread_id;
+       UNLOCK(_pthread_list_lock);
+       return rval;
+}
+#endif
+
+int
+pthread_getname_np(pthread_t thread, char *threadname, size_t len)
+{
+       int rval;
+       rval = 0;
+
+       if (thread == NULL)
+               return(ESRCH);
+       
+       LOCK(_pthread_list_lock);
+       if ((rval = _pthread_find_thread(thread)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               return(rval);
+       }
+       strlcpy(threadname, thread->pthread_name, len);
+       UNLOCK(_pthread_list_lock);
+       return rval;
+}
+
+int
+pthread_setname_np(const char *threadname)
+{
+       int rval;
+       int len;
+
+       rval = 0;
+       len = strlen(threadname);
+
+       /* protytype is in pthread_internals.h */
+       rval = proc_setthreadname((void *)threadname, len);
+       if(rval == 0)
+       {
+               strlcpy((pthread_self())->pthread_name, threadname, MAXTHREADNAMESIZE);
+       }
+       return rval;
+
+}
+
 static int       
-_pthread_create_suspended(pthread_t *thread, 
+_new_pthread_create_suspended(pthread_t *thread, 
               const pthread_attr_t *attr,
               void *(*start_routine)(void *), 
               void *arg,
-           int suspended)
+               int create_susp)
 {
        pthread_attr_t *attrs;
        void *stack;
-       int res;
-       pthread_t t;
+       int error;
+       unsigned int flags;
+       pthread_t t,t2;
        kern_return_t kern_res;
        mach_port_t kernel_thread = MACH_PORT_NULL;
        int needresume;
+       task_t self = mach_task_self();
+       int kernalloc = 0;
+       int susp = create_susp;
 
        if ((attrs = (pthread_attr_t *)attr) == (pthread_attr_t *)NULL)
        {                       /* Set up default paramters */
@@ -702,62 +1210,244 @@ _pthread_create_suspended(pthread_t *thread,
        } else if (attrs->sig != _PTHREAD_ATTR_SIG) {
             return EINVAL;
        }
-       res = ESUCCESS;
+       error = 0;
 
-       /* In default policy (ie SCHED_OTHER) only sched_priority is used. Check for
-        * any change in priority or policy is needed here.
-        */
        if (((attrs->policy != _PTHREAD_DEFAULT_POLICY)  || 
-               (attrs->param.sched_priority != default_priority)) && (suspended == 0)) {
+               (attrs->param.sched_priority != default_priority)) && (create_susp == 0)) {
                needresume = 1; 
-               suspended = 1;  
+               susp = 1;       
        } else 
                needresume = 0;
 
-       do
-       {
+       /* In default policy (ie SCHED_OTHER) only sched_priority is used. Check for
+        * any change in priority or policy is needed here.
+        */
+       if ((__oldstyle == 1) || (create_susp != 0)) {
+               /* Rosetta or pthread_create_suspended() */
+               /* running under rosetta */
                /* Allocate a stack for the thread */
-                if ((res = _pthread_allocate_stack(attrs, &stack)) != 0) {
-                    break;
+#if PTH_TRACE
+               __kdebug_trace(0x9000000, create_susp, 0, 0, 0, 0);
+#endif
+                if ((error = _pthread_allocate_stack(attrs, &stack)) != 0) {
+                       return(error);
                 }
                t = (pthread_t)malloc(sizeof(struct _pthread));
                *thread = t;
-               if (suspended) {
+               if (susp) {
                        /* Create the Mach thread for this thread */
-                       PTHREAD_MACH_CALL(thread_create(mach_task_self(), &kernel_thread), kern_res);
+                       PTHREAD_MACH_CALL(thread_create(self, &kernel_thread), kern_res);
                        if (kern_res != KERN_SUCCESS)
                        {
                                printf("Can't create thread: %d\n", kern_res);
-                               res = EINVAL; /* Need better error here? */
-                               break;
+                               return(EINVAL);
                        }
                }
-                if ((res = _pthread_create(t, attrs, stack, kernel_thread)) != 0)
+                if ((error = _pthread_create(t, attrs, stack, kernel_thread)) != 0)
                {
-                       break;
+                       return(error);
                }
                set_malloc_singlethreaded(0);
                __is_threaded = 1;
-               LOCK(_pthread_count_lock);
-               _pthread_count++;
-               UNLOCK(_pthread_count_lock);
 
                /* Send it on it's way */
                t->arg = arg;
                t->fun = start_routine;
+               t->newstyle = 0;
                /* Now set it up to execute */
-               _pthread_setup(t, _pthread_body, stack, suspended, needresume);
-       } while (0);
-       return (res);
-}
+               LOCK(_pthread_list_lock);
+               TAILQ_INSERT_TAIL(&__pthread_head, t, plist);
+#if PTH_LISTTRACE
+               __kdebug_trace(0x900000c, t, 0, 0, 4, 0);
+#endif
+               _pthread_count++;
+               UNLOCK(_pthread_list_lock);
+               _pthread_setup(t, _pthread_body, stack, susp, needresume);
+               return(0);
+       } else {
 
-int
-pthread_create(pthread_t *thread,
-           const pthread_attr_t *attr,
-           void *(*start_routine)(void *),
+               flags = 0;
+               if (attrs->fastpath == 1)
+                       kernalloc = 1;
+
+               if (attrs->detached == PTHREAD_CREATE_DETACHED)
+                               flags |= PTHREAD_START_DETACHED;
+               if (attrs->schedset != 0) {
+                       flags |= PTHREAD_START_SETSCHED;
+                       flags |= ((attrs->policy & PTHREAD_START_POLICY_MASK) << PTHREAD_START_POLICY_BITSHIFT);
+                       flags |= (attrs->param.sched_priority & PTHREAD_START_IMPORTANCE_MASK);
+               } 
+
+               set_malloc_singlethreaded(0);
+               __is_threaded = 1;
+
+               if (kernalloc == 0) {
+                       /* Allocate a stack for the thread */
+                       flags |= PTHREAD_START_CUSTOM;
+                       if ((error = _pthread_create_pthread_onstack(attrs, &stack, &t)) != 0) {
+                               return(error);
+                       }
+                       /* Send it on it's way */
+                       t->arg = arg;
+                       t->fun = start_routine;
+                       t->newstyle = 1;
+
+#if PTH_TRACE
+               __kdebug_trace(0x9000004, t, flags, 0, 0, 0);
+#endif
+                       
+                       if ((t2 = __bsdthread_create(start_routine, arg, stack, t, flags)) == (pthread_t)-1) {
+                               _pthread_free_pthread_onstack(t, 1, 0);
+                               return (EAGAIN);
+                       }
+                       else t=t2;
+                       LOCK(_pthread_list_lock);
+                       t->parentcheck = 1;
+                       if ((t->childexit != 0) && ((t->detached & PTHREAD_CREATE_DETACHED) == PTHREAD_CREATE_DETACHED)) {
+                               /* detached child exited, mop up */
+                               UNLOCK(_pthread_list_lock);
+#if PTH_TRACE
+                       __kdebug_trace(0x9000008, t, 0, 0, 1, 0);
+#endif
+                       if(t->freeStackOnExit)
+                                vm_deallocate(self, (mach_vm_address_t)(uintptr_t)t, pthreadsize);
+                        else
+                               free(t);
+                       } else if (t->childrun == 0) {
+                               TAILQ_INSERT_TAIL(&__pthread_head, t, plist);
+                               _pthread_count++;
+#if PTH_LISTTRACE
+                               __kdebug_trace(0x900000c, t, 0, 0, 1, 0);
+#endif
+                               UNLOCK(_pthread_list_lock);
+                       } else 
+                               UNLOCK(_pthread_list_lock);
+
+                       *thread = t;
+
+#if PTH_TRACE
+               __kdebug_trace(0x9000014, t, 0, 0, 1, 0);
+#endif
+                       return (0);
+
+               } else {
+                       /* kernel allocation */
+#if PTH_TRACE
+               __kdebug_trace(0x9000018, flags, 0, 0, 0, 0);
+#endif
+                       if ((t = __bsdthread_create(start_routine, arg, (void *)attrs->stacksize, NULL, flags)) == (pthread_t)-1)
+                               return (EAGAIN);
+                       /* Now set it up to execute */
+                       LOCK(_pthread_list_lock);
+                       t->parentcheck = 1;
+                       if ((t->childexit != 0) && ((t->detached & PTHREAD_CREATE_DETACHED) == PTHREAD_CREATE_DETACHED)) {
+                               /* detached child exited, mop up */
+                               UNLOCK(_pthread_list_lock);
+#if PTH_TRACE
+                       __kdebug_trace(0x9000008, t, pthreadsize, 0, 2, 0);
+#endif
+                               vm_deallocate(self, (mach_vm_address_t)(uintptr_t)t, pthreadsize);
+                       } else if (t->childrun == 0) {
+                               TAILQ_INSERT_TAIL(&__pthread_head, t, plist);
+                               _pthread_count++;
+#if PTH_LISTTRACE
+                               __kdebug_trace(0x900000c, t, 0, 0, 2, 0);
+#endif
+                               UNLOCK(_pthread_list_lock);
+                       } else 
+                               UNLOCK(_pthread_list_lock);
+
+                       *thread = t;
+
+#if PTH_TRACE
+                       __kdebug_trace(0x9000014, t, 0, 0, 2, 0);
+#endif
+                       return(0);
+               }
+       }
+}
+
+static int       
+_pthread_create_suspended(pthread_t *thread, 
+              const pthread_attr_t *attr,
+              void *(*start_routine)(void *), 
+              void *arg,
+           int suspended)
+{
+       pthread_attr_t *attrs;
+       void *stack;
+       int res;
+       pthread_t t;
+       kern_return_t kern_res;
+       mach_port_t kernel_thread = MACH_PORT_NULL;
+       int needresume;
+
+       if ((attrs = (pthread_attr_t *)attr) == (pthread_attr_t *)NULL)
+       {                       /* Set up default paramters */
+               attrs = &_pthread_attr_default;
+       } else if (attrs->sig != _PTHREAD_ATTR_SIG) {
+            return EINVAL;
+       }
+       res = 0;
+
+       /* In default policy (ie SCHED_OTHER) only sched_priority is used. Check for
+        * any change in priority or policy is needed here.
+        */
+       if (((attrs->policy != _PTHREAD_DEFAULT_POLICY)  || 
+               (attrs->param.sched_priority != default_priority)) && (suspended == 0)) {
+               needresume = 1; 
+               suspended = 1;  
+       } else 
+               needresume = 0;
+
+       do
+       {
+               /* Allocate a stack for the thread */
+                if ((res = _pthread_allocate_stack(attrs, &stack)) != 0) {
+                    break;
+                }
+               t = (pthread_t)malloc(sizeof(struct _pthread));
+               *thread = t;
+               if (suspended) {
+                       /* Create the Mach thread for this thread */
+                       PTHREAD_MACH_CALL(thread_create(mach_task_self(), &kernel_thread), kern_res);
+                       if (kern_res != KERN_SUCCESS)
+                       {
+                               printf("Can't create thread: %d\n", kern_res);
+                               res = EINVAL; /* Need better error here? */
+                               break;
+                       }
+               }
+                if ((res = _pthread_create(t, attrs, stack, kernel_thread)) != 0)
+               {
+                       break;
+               }
+               set_malloc_singlethreaded(0);
+               __is_threaded = 1;
+
+               /* Send it on it's way */
+               t->arg = arg;
+               t->fun = start_routine;
+               /* Now set it up to execute */
+               LOCK(_pthread_list_lock);
+               TAILQ_INSERT_TAIL(&__pthread_head, t, plist);
+#if PTH_LISTTRACE
+               __kdebug_trace(0x900000c, t, 0, 0, 5, 0);
+#endif
+               _pthread_count++;
+               UNLOCK(_pthread_list_lock);
+               _pthread_setup(t, _pthread_body, stack, suspended, needresume);
+       } while (0);
+       return (res);
+}
+
+int
+pthread_create(pthread_t *thread,
+           const pthread_attr_t *attr,
+           void *(*start_routine)(void *),
            void *arg)
 {
-    return _pthread_create_suspended(thread, attr, start_routine, arg, 0);
+    return _new_pthread_create_suspended(thread, attr, start_routine, arg, 0);
 }
 
 int
@@ -775,16 +1465,22 @@ pthread_create_suspended_np(pthread_t *thread,
 int       
 pthread_detach(pthread_t thread)
 {
-       if (thread->sig == _PTHREAD_SIG)
+       int newstyle = 0;
+       int ret;
+
+       if ((ret = _pthread_lookup_thread(thread, NULL, 1)) != 0)
+               return (ret); /* Not a valid thread */
+               
+       LOCK(thread->lock);
+       newstyle = thread->newstyle;
+       if (thread->detached & PTHREAD_CREATE_JOINABLE)
        {
-               LOCK(thread->lock);
-               if (thread->detached & PTHREAD_CREATE_JOINABLE)
-               {
-                       if (thread->detached & _PTHREAD_EXITED) {
-                               UNLOCK(thread->lock);
-                               pthread_join(thread, NULL);
-                               return ESUCCESS;
-                       } else {
+               if (thread->detached & _PTHREAD_EXITED) {
+                       UNLOCK(thread->lock);
+                       pthread_join(thread, NULL);
+                       return 0;
+               } else {
+                       if (newstyle == 0) {
                                semaphore_t death = thread->death;
 
                                thread->detached &= ~PTHREAD_CREATE_JOINABLE;
@@ -792,14 +1488,22 @@ pthread_detach(pthread_t thread)
                                UNLOCK(thread->lock);
                                if (death)
                                        (void) semaphore_signal(death);
-                               return (ESUCCESS);
+                       } else {
+                               mach_port_t joinport = thread->joiner_notify;
+       
+                               thread->detached &= ~PTHREAD_CREATE_JOINABLE;
+                               thread->detached |= PTHREAD_CREATE_DETACHED;
+                               
+                               UNLOCK(thread->lock);
+                               if (joinport) {
+                                       semaphore_signal(joinport);
+                               }
                        }
-               } else {
-                       UNLOCK(thread->lock);
-                       return (EINVAL);
+                       return(0);
                }
        } else {
-               return (ESRCH); /* Not a valid thread */
+               UNLOCK(thread->lock);
+               return (EINVAL);
        }
 }
 
@@ -807,26 +1511,46 @@ pthread_detach(pthread_t thread)
 /* 
  * pthread_kill call to system call
  */
-
-
 int   
 pthread_kill (
         pthread_t th,
         int sig)
 {
        int error = 0;
+       mach_port_t kport = MACH_PORT_NULL;
        
        if ((sig < 0) || (sig > NSIG))
                return(EINVAL);
 
-       if (th && (th->sig == _PTHREAD_SIG)) {
-               error = __pthread_kill(pthread_mach_thread_np(th), sig);
-               if (error == -1)
-                       error = errno;
-               return(error);
+       if (_pthread_lookup_thread(th, &kport, 0) != 0)
+               return (ESRCH); /* Not a valid thread */
+
+       /* if the thread is a workqueue thread, just return error */
+       if ((th->wqthread != 0) && (th->wqkillset ==0)) {
+               return(ENOTSUP);
        }
+
+       error = __pthread_kill(kport, sig);
+
+       if (error == -1)
+               error = errno;
+       return(error);
+}
+
+int 
+__pthread_workqueue_setkill(int enable)
+{
+       pthread_t self = pthread_self();
+
+       LOCK(self->lock);
+       if (enable == 0)
+               self->wqkillset = 0;
        else
-               return(ESRCH);
+               self->wqkillset = 1;
+       UNLOCK(self->lock);
+
+       return(0);
+
 }
 
 /* Announce that there are pthread resources ready to be reclaimed in a */
@@ -834,7 +1558,7 @@ pthread_kill (
 /* thread underneath is terminated right away. */
 static
 void _pthread_become_available(pthread_t thread, mach_port_t kernel_thread) {
-       mach_msg_empty_rcv_t msg;
+       pthread_reap_msg_t msg;
        kern_return_t ret;
 
        msg.header.msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_MAKE_SEND,
@@ -842,14 +1566,15 @@ void _pthread_become_available(pthread_t thread, mach_port_t kernel_thread) {
        msg.header.msgh_size = sizeof msg - sizeof msg.trailer;
        msg.header.msgh_remote_port = thread_recycle_port;
        msg.header.msgh_local_port = kernel_thread; 
-       msg.header.msgh_id = (int)thread;
+       msg.header.msgh_id = 0x44454144; /* 'DEAD' */
+       msg.thread = thread;
        ret = mach_msg_send(&msg.header);
        assert(ret == MACH_MSG_SUCCESS);
 }
 
 /* Reap the resources for available threads */
-static
-int _pthread_reap_thread(pthread_t th, mach_port_t kernel_thread, void **value_ptr) {
+__private_extern__
+int _pthread_reap_thread(pthread_t th, mach_port_t kernel_thread, void **value_ptr, int conforming) {
        mach_port_type_t ptype;
        kern_return_t ret;
        task_t self;
@@ -883,11 +1608,9 @@ int _pthread_reap_thread(pthread_t th, mach_port_t kernel_thread, void **value_p
                vm_address_t addr = (vm_address_t)th->stackaddr;
                vm_size_t size;
 
-               size = (vm_size_t)th->stacksize + vm_page_size;
+               size = (vm_size_t)th->stacksize + th->guardsize;
                
-#if !defined(STACK_GROWS_UP)
                addr -= size;
-#endif
                ret = vm_deallocate(self, addr, size);
                if (ret != KERN_SUCCESS) {
                  fprintf(stderr,
@@ -896,36 +1619,52 @@ int _pthread_reap_thread(pthread_t th, mach_port_t kernel_thread, void **value_p
                }
        }
 
-       if (value_ptr)
+
+       if (value_ptr) 
                *value_ptr = th->exit_value;
+       if (conforming) {
+               if ((th->cancel_state & (PTHREAD_CANCEL_ENABLE|_PTHREAD_CANCEL_PENDING)) ==
+            (PTHREAD_CANCEL_ENABLE|_PTHREAD_CANCEL_PENDING) && (value_ptr != NULL)) 
+                       *value_ptr = PTHREAD_CANCELED;
+               th->sig = _PTHREAD_NO_SIG;
+       }
+
 
        if (th != &_thread)
                free(th);
 
-       return ESUCCESS;
+       return 0;
 }
 
 static
 void _pthread_reap_threads(void)
 {
-       mach_msg_empty_rcv_t msg;
+       pthread_reap_msg_t msg;
        kern_return_t ret;
 
        ret = mach_msg(&msg.header, MACH_RCV_MSG|MACH_RCV_TIMEOUT, 0,
-                       sizeof(mach_msg_empty_rcv_t), thread_recycle_port, 
+                       sizeof msg, thread_recycle_port, 
                        MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
        while (ret == MACH_MSG_SUCCESS) {
                mach_port_t kernel_thread = msg.header.msgh_remote_port;
-               pthread_t thread = (pthread_t)msg.header.msgh_id;
+               pthread_t thread = msg.thread;
 
-               if (_pthread_reap_thread(thread, kernel_thread, (void **)0) == EAGAIN)
+               /* deal with race with thread_create_running() */
+               if (kernel_thread == MACH_PORT_NULL &&
+                   kernel_thread != thread->kernel_thread) {
+                       kernel_thread = thread->kernel_thread;
+               }
+               
+               if ( kernel_thread == MACH_PORT_NULL ||
+                    _pthread_reap_thread(thread, kernel_thread, (void **)0, 0) == EAGAIN)
                {
                        /* not dead yet, put it back for someone else to reap, stop here */
                        _pthread_become_available(thread, kernel_thread);
                        return;
                }
+
                ret = mach_msg(&msg.header, MACH_RCV_MSG|MACH_RCV_TIMEOUT, 0,
-                               sizeof(mach_msg_empty_rcv_t), thread_recycle_port, 
+                               sizeof msg, thread_recycle_port, 
                                MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
        }
 }
@@ -940,116 +1679,113 @@ _pthread_self() {
 /*
  * Terminate a thread.
  */
-void 
-pthread_exit(void *value_ptr)
+int __disable_threadsignal(int);
+
+static void 
+_pthread_exit(pthread_t self, void *value_ptr)
 {
-       struct _pthread_handler_rec *handler;
-       pthread_t self = pthread_self();
+       struct __darwin_pthread_handler_rec *handler;
        kern_return_t kern_res;
        int thread_count;
+       int newstyle = self->newstyle;
 
        /* Make this thread not to receive any signals */
-       syscall(331,1);
+       __disable_threadsignal(1);
+
+#if PTH_TRACE
+       __kdebug_trace(0x900001c, self, newstyle, 0, 0, 0);
+#endif
+
+       /* set cancel state to disable and type to deferred */
+       _pthread_setcancelstate_exit(self, value_ptr, __unix_conforming);
 
-       while ((handler = self->cleanup_stack) != 0)
+       while ((handler = self->__cleanup_stack) != 0)
        {
-               (handler->routine)(handler->arg);
-               self->cleanup_stack = handler->next;
+               (handler->__routine)(handler->__arg);
+               self->__cleanup_stack = handler->__next;
        }
        _pthread_tsd_cleanup(self);
 
-       _pthread_reap_threads();
-
-       LOCK(self->lock);
-       self->detached |= _PTHREAD_EXITED;
-
-       if (self->detached & PTHREAD_CREATE_JOINABLE) {
-               mach_port_t death = self->death;
-               self->exit_value = value_ptr;
-               UNLOCK(self->lock);
-               /* the joiner will need a kernel thread reference, leave ours for it */
-               if (death) {
-                       PTHREAD_MACH_CALL(semaphore_signal(death), kern_res);
-                       if (kern_res != KERN_SUCCESS)
-                               fprintf(stderr,
-                                   "semaphore_signal(death) failed: %s\n",
-                                       mach_error_string(kern_res));
+       if (newstyle == 0) {
+               _pthread_reap_threads();
+
+               LOCK(self->lock);
+               self->detached |= _PTHREAD_EXITED;
+
+               if (self->detached & PTHREAD_CREATE_JOINABLE) {
+                       mach_port_t death = self->death;
+                       self->exit_value = value_ptr;
+                       UNLOCK(self->lock);
+                       /* the joiner will need a kernel thread reference, leave ours for it */
+                       if (death) {
+                               PTHREAD_MACH_CALL(semaphore_signal(death), kern_res);
+                               if (kern_res != KERN_SUCCESS)
+                                       fprintf(stderr,
+                                               "semaphore_signal(death) failed: %s\n",
+                                               mach_error_string(kern_res));
+                       }
+                       LOCK(_pthread_list_lock);
+                       thread_count = --_pthread_count;
+                       UNLOCK(_pthread_list_lock);
+               } else {
+                       UNLOCK(self->lock);
+                       LOCK(_pthread_list_lock);
+                       TAILQ_REMOVE(&__pthread_head, self, plist);
+#if PTH_LISTTRACE
+                       __kdebug_trace(0x9000010, self, 0, 0, 5, 0);
+#endif
+                       thread_count = --_pthread_count;
+                       UNLOCK(_pthread_list_lock);
+                       /* with no joiner, we let become available consume our cached ref */
+                       _pthread_become_available(self, self->kernel_thread);
                }
-       } else {
-               UNLOCK(self->lock);
-               /* with no joiner, we let become available consume our cached ref */
-               _pthread_become_available(self, pthread_mach_thread_np(self));
-       }
 
-       LOCK(_pthread_count_lock);
-       thread_count = --_pthread_count;
-       UNLOCK(_pthread_count_lock);
-       if (thread_count <= 0)
-               exit(0);
+               if (thread_count <= 0)
+                       exit(0);
 
-       /* Use a new reference to terminate ourselves. Should never return. */
-       PTHREAD_MACH_CALL(thread_terminate(mach_thread_self()), kern_res);
-       fprintf(stderr, "thread_terminate(mach_thread_self()) failed: %s\n",
-                       mach_error_string(kern_res));
-       abort();
-}
-
-/*
- * Wait for a thread to terminate and obtain its exit value.
- */
-int       
-pthread_join(pthread_t thread, 
-            void **value_ptr)
-{
-       kern_return_t kern_res;
-       int res = ESUCCESS;
-
-       if (thread->sig == _PTHREAD_SIG)
-       {
-               semaphore_t death = new_sem_from_pool(); /* in case we need it */
-
-               LOCK(thread->lock);
-               if ((thread->detached & PTHREAD_CREATE_JOINABLE) &&
-                       thread->death == SEMAPHORE_NULL)
-               {
-                       pthread_t self = pthread_self();
-
-                       assert(thread->joiner == NULL);
-                       if (thread != self && (self == NULL || self->joiner != thread))
-                       {
-                               int already_exited = (thread->detached & _PTHREAD_EXITED);
-
-                               thread->death = death;
-                               thread->joiner = self;
-                               UNLOCK(thread->lock);
+               /* Use a new reference to terminate ourselves. Should never return. */
+               PTHREAD_MACH_CALL(thread_terminate(mach_thread_self()), kern_res);
+               fprintf(stderr, "thread_terminate(mach_thread_self()) failed: %s\n",
+                               mach_error_string(kern_res));
+       } else {
+               semaphore_t joinsem = SEMAPHORE_NULL;
 
-                               if (!already_exited)
-                               {
-                                       /* Wait for it to signal... */ 
-                                       do {
-                                               PTHREAD_MACH_CALL(semaphore_wait(death), kern_res);
-                                       } while (kern_res != KERN_SUCCESS);
-                               }
+               if ((self->joiner_notify == (mach_port_t)0) && (self->detached & PTHREAD_CREATE_JOINABLE))
+                       joinsem = new_sem_from_pool();
+               LOCK(self->lock);
+               self->detached |= _PTHREAD_EXITED;
 
-                               /* ... and wait for it to really be dead */
-                               while ((res = _pthread_reap_thread(thread,
-                                                       thread->kernel_thread,
-                                                       value_ptr)) == EAGAIN)
-                               {
-                                       sched_yield();
-                               }
-                       } else {
-                               UNLOCK(thread->lock);
-                               res = EDEADLK;
+               self->exit_value = value_ptr;
+               if (self->detached & PTHREAD_CREATE_JOINABLE) {
+                       if (self->joiner_notify == (mach_port_t)0) {
+                               self->joiner_notify = joinsem;
+                               joinsem = SEMAPHORE_NULL;
                        }
+                       UNLOCK(self->lock);
+                       if (joinsem != SEMAPHORE_NULL)
+                               restore_sem_to_pool(joinsem);
+                       _pthread_free_pthread_onstack(self, 0, 1);
                } else {
-                       UNLOCK(thread->lock);
-                       res = EINVAL;
+                       UNLOCK(self->lock);
+                       /* with no joiner, we let become available consume our cached ref */
+                       if (joinsem != SEMAPHORE_NULL)
+                               restore_sem_to_pool(joinsem);
+                       _pthread_free_pthread_onstack(self, 1, 1);
                }
-               restore_sem_to_pool(death);
-               return res;
        }
-       return ESRCH;
+       LIBC_ABORT("thread %p didn't exit", self);
+}
+
+void
+pthread_exit(void *value_ptr)
+{
+       pthread_t self = pthread_self();
+       /* if the current thread is a workqueue thread, just crash the app, as per libdispatch folks */
+       if (self->wqthread == 0) {
+               _pthread_exit(self, value_ptr);
+       } else {
+               LIBC_ABORT("pthread_exit() may only be called against threads created via pthread_create()");
+       }
 }
 
 /*
@@ -1060,22 +1796,32 @@ pthread_getschedparam(pthread_t thread,
                      int *policy,
                      struct sched_param *param)
 {
-       if (thread->sig == _PTHREAD_SIG)
-       {
-            *policy = thread->policy;
-            *param = thread->param;
-            return (ESUCCESS);
-       } else
-       {
-               return (ESRCH);  /* Not a valid thread structure */
+       int ret;
+
+       if (thread == NULL)
+               return(ESRCH);
+       
+       LOCK(_pthread_list_lock);
+
+       if ((ret = _pthread_find_thread(thread)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               return(ret);
        }
+       if (policy != 0)
+               *policy = thread->policy;
+       if (param != 0)
+               *param = thread->param;
+       UNLOCK(_pthread_list_lock);
+
+       return(0);
 }
 
 /*
  * Set the scheduling policy and scheduling paramters for a thread.
  */
-int       
-pthread_setschedparam(pthread_t thread, 
+static int       
+pthread_setschedparam_internal(pthread_t thread, 
+                     mach_port_t  kport,
                      int policy,
                      const struct sched_param *param)
 {
@@ -1084,10 +1830,8 @@ pthread_setschedparam(pthread_t thread,
        mach_msg_type_number_t count;
        kern_return_t ret;
 
-       if (thread->sig == _PTHREAD_SIG)
+       switch (policy)
        {
-               switch (policy)
-               {
                case SCHED_OTHER:
                        bases.ts.base_priority = param->sched_priority;
                        base = (policy_base_t)&bases.ts;
@@ -1107,19 +1851,47 @@ pthread_setschedparam(pthread_t thread,
                        break;
                default:
                        return (EINVAL);
-               }
-            thread->policy = policy;
-            thread->param = *param;
-               ret = thread_policy(pthread_mach_thread_np(thread), policy, base, count, TRUE);
-               if (ret != KERN_SUCCESS)
-               {
+       }
+       ret = thread_policy(kport, policy, base, count, TRUE);
+       if (ret != KERN_SUCCESS)
                        return (EINVAL);
-               }
-               return (ESUCCESS);
+       return (0);
+}
+
+int       
+pthread_setschedparam(pthread_t t, 
+                     int policy,
+                     const struct sched_param *param)
+{
+       mach_port_t kport = MACH_PORT_NULL;
+       int error;
+       int bypass = 1;
+
+       if (t != pthread_self() && t != &_thread ) { //since the main thread will not get de-allocated from underneath us
+               bypass = 0;
+               if (_pthread_lookup_thread(t, &kport, 0) != 0)
+                       return(ESRCH);
        } else
-       {
-               return (ESRCH);  /* Not a valid thread structure */
+               kport = t->kernel_thread;
+
+       error = pthread_setschedparam_internal(t, kport, policy, param);
+       if (error == 0) {
+               if (bypass == 0) {
+                       /* ensure the thread is still valid */
+                       LOCK(_pthread_list_lock);
+                       if ((error = _pthread_find_thread(t)) != 0) {
+                               UNLOCK(_pthread_list_lock);
+                               return(error);
+                       }
+                       t->policy = policy;
+                       t->param = *param;
+                       UNLOCK(_pthread_list_lock);
+               }  else {
+                       t->policy = policy;
+                       t->param = *param;
+               }
        }
+       return(error);
 }
 
 /*
@@ -1150,6 +1922,23 @@ pthread_equal(pthread_t t1,
        return (t1 == t2);
 }
 
+// Force LLVM not to optimise this to a call to __pthread_set_self, if it does
+// then _pthread_set_self won't be bound when secondary threads try and start up.
+void __attribute__((noinline))
+_pthread_set_self(pthread_t p)
+{
+       extern void __pthread_set_self(void *);
+
+        if (p == 0) {
+                if (_thread.tsd[0] != 0) {
+                        bzero(&_thread, sizeof(struct _pthread));
+                }
+                p = &_thread;
+        }
+        p->tsd[0] = p;
+       __pthread_set_self(&p->tsd[0]);
+}
+
 void 
 cthread_set_self(void *cself)
 {
@@ -1170,6 +1959,17 @@ ur_cthread_self(void) {
     return self->cthread_self;
 }
 
+/*
+ * cancellation handler for pthread once as the init routine can have a
+ * cancellation point. In that case we need to restore the spin unlock 
+ */
+void
+__pthread_once_cancel_handler(pthread_once_t *once_control)
+{
+               _spin_unlock(&once_control->lock);
+}
+
+
 /*
  * Execute a function exactly one time in a thread-safe fashion.
  */
@@ -1177,135 +1977,75 @@ int
 pthread_once(pthread_once_t *once_control, 
             void (*init_routine)(void))
 {
-       LOCK(once_control->lock);
+       _spin_lock(&once_control->lock);
        if (once_control->sig == _PTHREAD_ONCE_SIG_init)
        {
+               pthread_cleanup_push((void (*)(void *))__pthread_once_cancel_handler, once_control);
                (*init_routine)();
+               pthread_cleanup_pop(0);
                once_control->sig = _PTHREAD_ONCE_SIG;
        }
-       UNLOCK(once_control->lock);
-       return (ESUCCESS);  /* Spec defines no possible errors! */
-}
-
-/*
- * Cancel a thread
- */
-int
-pthread_cancel(pthread_t thread)
-{
-       if (thread->sig == _PTHREAD_SIG)
-       {
-               thread->cancel_state |= _PTHREAD_CANCEL_PENDING;
-               return (ESUCCESS);
-       } else
-       {
-               return (ESRCH);
-       }
+       _spin_unlock(&once_control->lock);
+       return (0);  /* Spec defines no possible errors! */
 }
 
 /*
  * Insert a cancellation point in a thread.
  */
-static void
-_pthread_testcancel(pthread_t thread)
+__private_extern__ void
+_pthread_testcancel(pthread_t thread, int isconforming)
 {
        LOCK(thread->lock);
        if ((thread->cancel_state & (PTHREAD_CANCEL_ENABLE|_PTHREAD_CANCEL_PENDING)) == 
            (PTHREAD_CANCEL_ENABLE|_PTHREAD_CANCEL_PENDING))
        {
                UNLOCK(thread->lock);
-               pthread_exit(0);
+               if (isconforming)
+                       pthread_exit(PTHREAD_CANCELED);
+               else
+                       pthread_exit(0);
        }
        UNLOCK(thread->lock);
 }
 
-void
-pthread_testcancel(void)
+
+
+int
+pthread_getconcurrency(void)
 {
-       pthread_t self = pthread_self();
-       _pthread_testcancel(self);
-}
-
-/*
- * Query/update the cancelability 'state' of a thread
- */
-int
-pthread_setcancelstate(int state, int *oldstate)
-{
-       pthread_t self = pthread_self();
-       int err = ESUCCESS;
-       LOCK(self->lock);
-       if (oldstate)
-               *oldstate = self->cancel_state & ~_PTHREAD_CANCEL_STATE_MASK;
-       if ((state == PTHREAD_CANCEL_ENABLE) || (state == PTHREAD_CANCEL_DISABLE))
-       {
-               self->cancel_state = (self->cancel_state & _PTHREAD_CANCEL_STATE_MASK) | state;
-       } else
-       {
-               err = EINVAL;
-       }
-       UNLOCK(self->lock);
-       _pthread_testcancel(self);  /* See if we need to 'die' now... */
-       return (err);
-}
-
-/*
- * Query/update the cancelability 'type' of a thread
- */
-int
-pthread_setcanceltype(int type, int *oldtype)
-{
-       pthread_t self = pthread_self();
-       int err = ESUCCESS;
-       LOCK(self->lock);
-       if (oldtype)
-               *oldtype = self->cancel_state & ~_PTHREAD_CANCEL_TYPE_MASK;
-       if ((type == PTHREAD_CANCEL_DEFERRED) || (type == PTHREAD_CANCEL_ASYNCHRONOUS))
-       {
-               self->cancel_state = (self->cancel_state & _PTHREAD_CANCEL_TYPE_MASK) | type;
-       } else
-       {
-               err = EINVAL;
-       }
-       UNLOCK(self->lock);
-       _pthread_testcancel(self);  /* See if we need to 'die' now... */
-       return (err);
-}
-
-int
-pthread_getconcurrency(void)
-{
-       return(pthread_concurrency);
+       return(pthread_concurrency);
 }
 
 int
 pthread_setconcurrency(int new_level)
 {
+       if (new_level < 0)
+               return EINVAL;
        pthread_concurrency = new_level;
-       return(ESUCCESS);
+       return(0);
 }
 
 /*
  * Perform package initialization - called automatically when application starts
  */
-
-static int
+int
 pthread_init(void)
 {
        pthread_attr_t *attrs;
         pthread_t thread;
        kern_return_t kr;
-       host_basic_info_data_t basic_info;
        host_priority_info_data_t priority_info;
        host_info_t info;
        host_flavor_t flavor;
        host_t host;
        mach_msg_type_number_t count;
        int mib[2];
+       int ncpus = 0;
        size_t len;
-       int numcpus;
+       void *stackaddr;
 
-        count = HOST_PRIORITY_INFO_COUNT;
+       pthreadsize = round_page(sizeof (struct _pthread));
+       count = HOST_PRIORITY_INFO_COUNT;
        info = (host_info_t)&priority_info;
        flavor = HOST_PRIORITY_INFO;
        host = mach_host_self();
@@ -1320,50 +2060,72 @@ pthread_init(void)
        attrs = &_pthread_attr_default;
        pthread_attr_init(attrs);
 
+       TAILQ_INIT(&__pthread_head);
+       LOCK_INIT(_pthread_list_lock);
        thread = &_thread;
+       TAILQ_INSERT_HEAD(&__pthread_head, thread, plist);
        _pthread_set_self(thread);
-       _pthread_create(thread, attrs, (void *)USRSTACK, mach_thread_self());
+#if PTH_LISTTRACE
+       __kdebug_trace(0x900000c, thread, 0, 0, 10, 0);
+#endif
+       
+       /* In case of dyld reset the tsd keys from 1 - 10 */
+       _pthread_keys_init();
+
+       mib[0] = CTL_KERN;
+       mib[1] = KERN_USRSTACK;
+       len = sizeof (stackaddr);
+       if (sysctl (mib, 2, &stackaddr, &len, NULL, 0) != 0)
+                       stackaddr = (void *)USRSTACK;
+       _pthread_create(thread, attrs, stackaddr, mach_thread_self());
+       thread->stacksize = DFLSSIZ; //initialize main thread's stacksize based on vmparam.h
        thread->detached = PTHREAD_CREATE_JOINABLE|_PTHREAD_CREATE_PARENT;
 
-        /* See if we're on a multiprocessor and set _spin_tries if so.  */
-       mib[0] = CTL_HW;
-       mib[1] = HW_NCPU;
-       len = sizeof(numcpus);
-       if (sysctl(mib, 2, &numcpus, &len, NULL, 0) == 0) {
-               if (numcpus > 1) {
-                       _spin_tries = MP_SPIN_TRIES;
-               }
-       } else {
-               count = HOST_BASIC_INFO_COUNT;
-               info = (host_info_t)&basic_info;
-               flavor = HOST_BASIC_INFO;
-               kr = host_info(host, flavor, info, &count);
-               if (kr != KERN_SUCCESS)
-                       printf("host_info failed (%d)\n", kr);
-               else {
-                       if (basic_info.avail_cpus > 1)
-                               _spin_tries = MP_SPIN_TRIES;
-               }
-       }
+       _init_cpu_capabilities();
+       if ((ncpus = _NumCPUs()) > 1)
+               _spin_tries = MP_SPIN_TRIES;
+
+       workq_targetconc[WORKQ_HIGH_PRIOQUEUE] = ncpus;
+       workq_targetconc[WORKQ_DEFAULT_PRIOQUEUE] = ncpus;
+       workq_targetconc[WORKQ_LOW_PRIOQUEUE] =  ncpus;
+       workq_targetconc[WORKQ_BG_PRIOQUEUE] =  ncpus;
 
        mach_port_deallocate(mach_task_self(), host);
     
-       _init_cpu_capabilities();       /* check for vector unit, cache line size etc */
-
 #if defined(__ppc__)
-       /* Use fsqrt instruction in sqrt() if available. */
-    if (_cpu_capabilities & kHasFsqrt) {
-        extern size_t hw_sqrt_len;
-        extern double sqrt( double );
-        extern double hw_sqrt( double );
-        extern void sys_icache_invalidate(void *, size_t);
-
-        memcpy ( (void *)sqrt, (void *)hw_sqrt, hw_sqrt_len );
-        sys_icache_invalidate((void *)sqrt, hw_sqrt_len);
-    }
+       IF_ROSETTA() {
+               __oldstyle = 1;
+       }
 #endif
-    
+
+#if defined(_OBJC_PAGE_BASE_ADDRESS)
+{
+        vm_address_t objcRTPage = (vm_address_t)_OBJC_PAGE_BASE_ADDRESS;
+        kr = vm_map(mach_task_self(),
+                 &objcRTPage, vm_page_size * 4, vm_page_size - 1,
+                 VM_FLAGS_FIXED | VM_MAKE_TAG(0), // Which tag to use?
+                 MACH_PORT_NULL,
+                 (vm_address_t)0, FALSE,
+                 (vm_prot_t)0, VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE,
+                 VM_INHERIT_DEFAULT);
+               /* We ignore the return result here. The ObjC runtime will just have to deal. */
+}
+#endif
+       //added so that thread_recycle_port is initialized on new launch.
+       _pthread_fork_child_postinit();
        mig_init(1);            /* enable multi-threaded mig interfaces */
+       if (__oldstyle == 0) {
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+               __bsdthread_register(thread_start, start_wqthread, round_page(sizeof(struct _pthread)), _pthread_start, &workq_targetconc[0], (uintptr_t)(&thread->tsd[__PTK_LIBDISPATCH_KEY0]) - (uintptr_t)(&thread->tsd[0]));
+#else
+               __bsdthread_register(_pthread_start, _pthread_wqthread, round_page(sizeof(struct _pthread)), NULL, &workq_targetconc[0], (uintptr_t)&thread->tsd[__PTK_LIBDISPATCH_KEY0] - (uintptr_t)thread);
+#endif
+       }
+
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+       if( (thread->thread_id = __thread_selfid()) == (__uint64_t)-1)
+               printf("Failed to set thread_id in pthread_init\n");
+#endif
        return 0;
 }
 
@@ -1373,8 +2135,13 @@ int sched_yield(void)
     return 0;
 }
 
-/* This is the "magic" that gets the initialization routine called when the application starts */
-int (*_cthread_init_routine)(void) = pthread_init;
+/* This used to be the "magic" that gets the initialization routine called when the application starts */
+/*
+ * (These has been moved to setenv.c, so we can use it to fix a less than 10.5
+ * crt1.o issue)
+ * static int _do_nothing(void) { return 0; }
+ * int (*_cthread_init_routine)(void) = _do_nothing;
+ */
 
 /* Get a semaphore from the pool, growing it if necessary */
 
@@ -1411,11 +2178,1262 @@ static void sem_pool_reset(void) {
        UNLOCK(sem_pool_lock);
 }
 
-__private_extern__ void _pthread_fork_child(void) {
+__private_extern__ void _pthread_fork_child(pthread_t p) {
        /* Just in case somebody had it locked... */
        UNLOCK(sem_pool_lock);
        sem_pool_reset();
-       UNLOCK(_pthread_count_lock);
+       /* No need to hold the pthread_list_lock as no one other than this 
+        * thread is present at this time
+        */
+       TAILQ_INIT(&__pthread_head);
+       LOCK_INIT(_pthread_list_lock);
+       TAILQ_INSERT_HEAD(&__pthread_head, p, plist);
+#if PTH_LISTTRACE
+       __kdebug_trace(0x900000c, p, 0, 0, 10, 0);
+#endif
        _pthread_count = 1;
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+       if( (p->thread_id = __thread_selfid()) == (__uint64_t)-1)
+               printf("Failed to set thread_id in pthread_fork_child\n");
+#endif
+}
+
+void _pthread_fork_child_postinit() {
+       kern_return_t kr;
+
+       kr = mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_RECEIVE, &thread_recycle_port);
+       if (kr != KERN_SUCCESS) {
+               abort();
+       }
+}
+
+/*
+ * Query/update the cancelability 'state' of a thread
+ */
+int
+_pthread_setcancelstate_internal(int state, int *oldstate, int conforming)
+{
+       pthread_t self = pthread_self();
+
+
+       switch (state) {
+               case PTHREAD_CANCEL_ENABLE:
+               if (conforming)
+                       __pthread_canceled(1);
+                       break;
+               case PTHREAD_CANCEL_DISABLE:
+               if (conforming)
+                       __pthread_canceled(2);
+                       break;
+               default:
+                       return EINVAL;
+       }
+
+       self = pthread_self();
+       LOCK(self->lock);
+       if (oldstate)
+               *oldstate = self->cancel_state & _PTHREAD_CANCEL_STATE_MASK;
+       self->cancel_state &= ~_PTHREAD_CANCEL_STATE_MASK;
+       self->cancel_state |= state;
+       UNLOCK(self->lock);
+       if (!conforming)
+               _pthread_testcancel(self, 0);  /* See if we need to 'die' now... */
+       return (0);
+}
+
+/* When a thread exits set the cancellation state to DISABLE and DEFERRED */
+static void
+_pthread_setcancelstate_exit(pthread_t self, void * value_ptr, int conforming)
+{
+       LOCK(self->lock);
+       self->cancel_state &= ~(_PTHREAD_CANCEL_STATE_MASK | _PTHREAD_CANCEL_TYPE_MASK);
+       self->cancel_state |= (PTHREAD_CANCEL_DISABLE | PTHREAD_CANCEL_DEFERRED);
+       if ((value_ptr == PTHREAD_CANCELED)) {
+// 4597450: begin
+               self->detached |= _PTHREAD_WASCANCEL;
+// 4597450: end
+       }
+       UNLOCK(self->lock);
+}
+
+int
+_pthread_join_cleanup(pthread_t thread, void ** value_ptr, int conforming)
+{
+       kern_return_t res;
+       int ret;
+
+#if PTH_TRACE
+       __kdebug_trace(0x9000028, thread, 0, 0, 1, 0);
+#endif
+       /* The scenario where the joiner was waiting for the thread and
+        * the pthread detach happened on that thread. Then the semaphore
+        * will trigger but by the time joiner runs, the target thread could be
+        * freed. So we need to make sure that the thread is still in the list 
+        * and is joinable  before we continue with the join.
+        */
+       LOCK(_pthread_list_lock);
+       if ((ret = _pthread_find_thread(thread)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               /* returns ESRCH */
+               return(ret);
+       }
+       if ((thread->detached & PTHREAD_CREATE_JOINABLE) == 0) {
+               /* the thread might be a detached thread */
+               UNLOCK(_pthread_list_lock);
+               return(ESRCH);
+
+       }
+       /* It is still a joinable thread and needs to be reaped */
+       TAILQ_REMOVE(&__pthread_head, thread, plist);
+#if PTH_LISTTRACE
+       __kdebug_trace(0x9000010, thread, 0, 0, 3, 0);
+#endif
+       UNLOCK(_pthread_list_lock);
+
+       if (value_ptr)
+               *value_ptr = thread->exit_value;
+       if (conforming) {
+               if ((thread->cancel_state & (PTHREAD_CANCEL_ENABLE|_PTHREAD_CANCEL_PENDING)) ==
+                       (PTHREAD_CANCEL_ENABLE|_PTHREAD_CANCEL_PENDING) && (value_ptr != NULL)) {
+                       *value_ptr = PTHREAD_CANCELED;
+               }
+       }
+       if (thread->reply_port != MACH_PORT_NULL) {
+               res = mach_port_mod_refs(mach_task_self(), thread->reply_port, MACH_PORT_RIGHT_RECEIVE, -1);
+               if (res != KERN_SUCCESS) 
+                       fprintf(stderr,"mach_port_mod_refs(reply_port) failed: %s\n",mach_error_string(res));
+               thread->reply_port = MACH_PORT_NULL;
+       }
+       if (thread->freeStackOnExit) {
+               thread->sig = _PTHREAD_NO_SIG;
+#if PTH_TRACE
+       __kdebug_trace(0x9000028, thread, 0, 0, 2, 0);
+#endif
+               vm_deallocate(mach_task_self(), (mach_vm_address_t)(uintptr_t)thread, pthreadsize);
+       } else {
+               thread->sig = _PTHREAD_NO_SIG;
+#if PTH_TRACE
+       __kdebug_trace(0x9000028, thread, 0, 0, 3, 0);
+#endif
+               free(thread);
+       }
+       return(0);
+}
+
+/* ALWAYS called with list lock and return with list lock */
+int
+_pthread_find_thread(pthread_t thread)
+{
+       pthread_t p;
+
+loop:
+       TAILQ_FOREACH(p, &__pthread_head, plist) {
+               if (p == thread) {
+                       if (thread->kernel_thread == MACH_PORT_NULL) {
+                               UNLOCK(_pthread_list_lock);
+                               sched_yield();
+                               LOCK(_pthread_list_lock);
+                               goto loop;
+                       } 
+                       return(0);
+               }
+       }
+       return(ESRCH);
+}
+
+int
+_pthread_lookup_thread(pthread_t thread, mach_port_t * portp, int only_joinable)
+{
+       mach_port_t kport;
+       int ret = 0;
+
+       if (thread == NULL)
+               return(ESRCH);
+       
+       LOCK(_pthread_list_lock);
+
+       if ((ret = _pthread_find_thread(thread)) != 0) {
+               UNLOCK(_pthread_list_lock);
+               return(ret);
+       }
+       if ((only_joinable != 0) && ((thread->detached & PTHREAD_CREATE_DETACHED) != 0)) {
+               UNLOCK(_pthread_list_lock);
+               return(EINVAL);
+       } 
+       kport = thread->kernel_thread;
+       UNLOCK(_pthread_list_lock);
+       if (portp != NULL)
+               *portp = kport;
+       return(0);
+}
+
+/* XXXXXXXXXXXXX Pthread Workqueue Attributes XXXXXXXXXXXXXXXXXX */
+int 
+pthread_workqueue_attr_init_np(pthread_workqueue_attr_t * attrp)
+{
+       attrp->queueprio = WORKQ_DEFAULT_PRIOQUEUE;
+       attrp->sig = PTHREAD_WORKQUEUE_ATTR_SIG;
+       attrp->overcommit = 0;
+       return(0);
+}
+
+int 
+pthread_workqueue_attr_destroy_np(pthread_workqueue_attr_t * attr)
+{
+       if (attr->sig == PTHREAD_WORKQUEUE_ATTR_SIG)
+       {
+               return (0);
+       } else
+       {
+               return (EINVAL); /* Not an attribute structure! */
+       }
+}
+
+int 
+pthread_workqueue_attr_getqueuepriority_np(const pthread_workqueue_attr_t * attr, int * qpriop)
+{
+    if (attr->sig == PTHREAD_WORKQUEUE_ATTR_SIG) {
+        *qpriop = attr->queueprio;
+        return (0);
+    } else {
+        return (EINVAL); /* Not an attribute structure! */
+    }
+}
+
+
+int 
+pthread_workqueue_attr_setqueuepriority_np(pthread_workqueue_attr_t * attr, int qprio)
+{
+int error = 0;
+
+    if (attr->sig == PTHREAD_WORKQUEUE_ATTR_SIG) {
+               switch(qprio) {
+                       case WORKQ_HIGH_PRIOQUEUE:
+                       case WORKQ_DEFAULT_PRIOQUEUE:
+                       case WORKQ_LOW_PRIOQUEUE:
+                       case WORKQ_BG_PRIOQUEUE:
+                               attr->queueprio = qprio;
+                               break;
+                       default:
+                               error = EINVAL;
+               }
+    } else {
+       error = EINVAL;
+    }
+       return (error);
+}
+
+
+int 
+pthread_workqueue_attr_getovercommit_np(const pthread_workqueue_attr_t * attr, int * ocommp)
+{
+    if (attr->sig == PTHREAD_WORKQUEUE_ATTR_SIG) {
+        *ocommp = attr->overcommit;
+        return (0);
+    } else {
+        return (EINVAL); /* Not an attribute structure! */
+    }
+}
+
+
+int 
+pthread_workqueue_attr_setovercommit_np(pthread_workqueue_attr_t * attr, int ocomm)
+{
+int error = 0;
+
+    if (attr->sig == PTHREAD_WORKQUEUE_ATTR_SIG) {
+       attr->overcommit = ocomm;
+    } else {
+       error = EINVAL;
+    }
+       return (error);
+}
+/* XXXXXXXXXXXXX Pthread Workqueue support routines XXXXXXXXXXXXXXXXXX */
+
+static void
+workqueue_list_lock()
+{
+       OSSpinLockLock(&__workqueue_list_lock);
+}
+
+static void
+workqueue_list_unlock()
+{
+       OSSpinLockUnlock(&__workqueue_list_lock);
+}
+
+int
+pthread_workqueue_init_np()
+{
+       int ret;
+
+       workqueue_list_lock();
+       ret =_pthread_work_internal_init();
+       workqueue_list_unlock();
+
+       return(ret);
+}
+
+int
+pthread_workqueue_requestconcurrency_np(int queue, int request_concurrency)
+{
+       int error = 0;
+
+       if (queue < 0 || queue > WORKQ_NUM_PRIOQUEUE)
+               return(EINVAL);
+
+       error =__workq_kernreturn(WQOPS_THREAD_SETCONC, NULL, request_concurrency, queue);
+
+       if (error == -1)
+               return(errno);
+       return(0);
+}
+
+void
+pthread_workqueue_atfork_prepare(void)
+{
+       /* 
+        * NOTE:  Any workq additions here  
+        * should be for i386,x86_64 only 
+        */
+       dispatch_atfork_prepare();
+}
+
+void
+pthread_workqueue_atfork_parent(void)
+{
+       /* 
+        * NOTE:  Any workq additions here  
+        * should be for i386,x86_64 only 
+        */
+       dispatch_atfork_parent();
+}
+
+void
+pthread_workqueue_atfork_child(void)
+{
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+       /* 
+        * NOTE:  workq additions here  
+        * are for i386,x86_64 only as
+        * ppc and arm do not support it
+        */
+       __workqueue_list_lock = OS_SPINLOCK_INIT;
+       if (kernel_workq_setup != 0){
+          kernel_workq_setup = 0;
+          _pthread_work_internal_init();
+       }
+#endif
+       dispatch_atfork_child();
+}
+
+static int
+_pthread_work_internal_init(void)
+{
+       int i, error;
+       pthread_workqueue_head_t headp;
+       pthread_workqueue_t wq;
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+       pthread_t self = pthread_self();
+#endif
+
+       if (kernel_workq_setup == 0) {
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+               __bsdthread_register(thread_start, start_wqthread, round_page(sizeof(struct _pthread)), _pthread_start, &workq_targetconc[0], (uintptr_t)(&self->tsd[__PTK_LIBDISPATCH_KEY0]) - (uintptr_t)(&self->tsd[0]));
+#else
+               __bsdthread_register(_pthread_start, _pthread_wqthread, round_page(sizeof(struct _pthread)),NULL,NULL,0);
+#endif
+
+               _pthread_wq_attr_default.queueprio = WORKQ_DEFAULT_PRIOQUEUE;
+               _pthread_wq_attr_default.sig = PTHREAD_WORKQUEUE_ATTR_SIG;
+
+               for( i = 0; i< WORKQ_NUM_PRIOQUEUE; i++) {
+                       headp = __pthread_wq_head_tbl[i];
+                       TAILQ_INIT(&headp->wqhead);
+                       headp->next_workq = 0;
+               }
+
+               __workqueue_pool_ptr = NULL;
+               __workqueue_pool_size = round_page(sizeof(struct _pthread_workitem) * WORKITEM_POOL_SIZE);
+
+               __workqueue_pool_ptr = (struct _pthread_workitem *)mmap(NULL, __workqueue_pool_size,
+                       PROT_READ|PROT_WRITE,
+                       MAP_ANON | MAP_PRIVATE,
+                       0,
+                       0);
+
+               if (__workqueue_pool_ptr == MAP_FAILED) {
+                       /* Not expected to fail, if it does, always malloc for work items */
+                       __workqueue_nitems = WORKITEM_POOL_SIZE;
+                       __workqueue_pool_ptr = NULL;
+               } else 
+                       __workqueue_nitems = 0;
+
+               /* sets up the workitem pool */
+               grow_workitem();
+
+               /* since the size is less than a page, leaving this in malloc pool */
+               wq = (struct _pthread_workqueue *)malloc(sizeof(struct _pthread_workqueue) * WORKQUEUE_POOL_SIZE);
+               bzero(wq, (sizeof(struct _pthread_workqueue) * WORKQUEUE_POOL_SIZE));
+               for (i = 0; i < WORKQUEUE_POOL_SIZE; i++) {
+                       TAILQ_INSERT_TAIL(&__pthread_workqueue_pool_head, &wq[i], wq_list);
+               }
+
+               if (error = __workq_open()) {
+                       TAILQ_INIT(&__pthread_workitem_pool_head);
+                       TAILQ_INIT(&__pthread_workqueue_pool_head);
+                       if (__workqueue_pool_ptr != NULL) {
+                               munmap((void *)__workqueue_pool_ptr, __workqueue_pool_size);
+                       }
+                       free(wq);
+                       return(ENOMEM);
+               }
+               kernel_workq_setup = 1;
+       }
+       return(0);
+}
+
+
+/* This routine is called with list lock held */
+static pthread_workitem_t
+alloc_workitem(void)
+{
+       pthread_workitem_t witem;
+
+       if (TAILQ_EMPTY(&__pthread_workitem_pool_head)) {
+               /* the chunk size is set so some multiple of it is pool size */
+               if (__workqueue_nitems < WORKITEM_POOL_SIZE) {
+                       grow_workitem();
+               } else {
+                       workqueue_list_unlock();
+                       witem = malloc(sizeof(struct _pthread_workitem));
+                       workqueue_list_lock();
+                       witem->fromcache = 0;
+                       goto out;
+               }
+       }
+       witem = TAILQ_FIRST(&__pthread_workitem_pool_head);
+       TAILQ_REMOVE(&__pthread_workitem_pool_head, witem, item_entry);
+       witem->fromcache = 1;
+out:
+       witem->flags = 0;
+       witem->item_entry.tqe_next = 0;
+       witem->item_entry.tqe_prev = 0;
+       user_workitem_count++;
+       return(witem);
+}
+
+/* This routine is called with list lock held */
+static void
+free_workitem(pthread_workitem_t witem) 
+{
+       user_workitem_count--;
+       witem->flags = 0;
+       if (witem->fromcache != 0)
+               TAILQ_INSERT_TAIL(&__pthread_workitem_pool_head, witem, item_entry);
+       else
+               free(witem);
+}
+
+static void
+grow_workitem(void)
+{
+       pthread_workitem_t witemp;
+       int i;
+
+        witemp = &__workqueue_pool_ptr[__workqueue_nitems];
+       bzero(witemp, (sizeof(struct _pthread_workitem) * WORKITEM_CHUNK_SIZE));
+       for (i = 0; i < WORKITEM_CHUNK_SIZE; i++) {
+               witemp[i].fromcache = 1;
+               TAILQ_INSERT_TAIL(&__pthread_workitem_pool_head, &witemp[i], item_entry);
+       }
+       __workqueue_nitems += WORKITEM_CHUNK_SIZE;
+}
+
+/* This routine is called with list lock held */
+static pthread_workqueue_t
+alloc_workqueue(void)
+{
+       pthread_workqueue_t wq;
+
+       if (TAILQ_EMPTY(&__pthread_workqueue_pool_head)) {
+               workqueue_list_unlock();
+               wq = malloc(sizeof(struct _pthread_workqueue));
+               workqueue_list_lock();
+       } else {
+               wq = TAILQ_FIRST(&__pthread_workqueue_pool_head);
+               TAILQ_REMOVE(&__pthread_workqueue_pool_head, wq, wq_list);
+       }
+       user_workq_count++;
+       return(wq);
+}
+
+/* This routine is called with list lock held */
+static void
+free_workqueue(pthread_workqueue_t wq) 
+{
+       user_workq_count--;
+       TAILQ_INSERT_TAIL(&__pthread_workqueue_pool_head, wq, wq_list);
 }
 
+static void
+_pthread_workq_init(pthread_workqueue_t wq, const pthread_workqueue_attr_t * attr)
+{
+       bzero(wq, sizeof(struct _pthread_workqueue));
+       if (attr != NULL) {
+               wq->queueprio = attr->queueprio;
+               wq->overcommit = attr->overcommit;
+        } else {
+               wq->queueprio = WORKQ_DEFAULT_PRIOQUEUE;
+               wq->overcommit = 0;
+       }
+       LOCK_INIT(wq->lock);
+       wq->flags = 0;
+       TAILQ_INIT(&wq->item_listhead);
+       TAILQ_INIT(&wq->item_kernhead);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080ac, wq, &wq->item_listhead, wq->item_listhead.tqh_first,  wq->item_listhead.tqh_last, 0);
+#endif
+       wq->wq_list.tqe_next = 0;
+       wq->wq_list.tqe_prev = 0;
+       wq->sig = PTHREAD_WORKQUEUE_SIG;
+       wq->headp = __pthread_wq_head_tbl[wq->queueprio];
+}
+
+int 
+valid_workq(pthread_workqueue_t workq)
+{
+       if (workq->sig == PTHREAD_WORKQUEUE_SIG) 
+               return(1);
+       else
+               return(0);
+}
+
+
+/* called with list lock */
+static void
+pick_nextworkqueue_droplock()
+{
+       int i, curwqprio, val, found;
+       pthread_workqueue_head_t headp;
+       pthread_workqueue_t workq;
+       pthread_workqueue_t nworkq = NULL;
+
+#if WQ_TRACE
+       __kdebug_trace(0x9008098, kernel_workq_count, 0, 0,  0, 0);
+#endif
+loop:
+       while (kernel_workq_count < KERNEL_WORKQ_ELEM_MAX) {
+               found = 0;
+               for (i = 0; i < WORKQ_NUM_PRIOQUEUE; i++)  {
+                       wqreadyprio = i;        /* because there is nothing else higher to run */
+                       headp = __pthread_wq_head_tbl[i];
+
+                       if (TAILQ_EMPTY(&headp->wqhead))
+                               continue;
+                       workq = headp->next_workq;
+                       if (workq == NULL)
+                               workq = TAILQ_FIRST(&headp->wqhead);
+                       curwqprio = workq->queueprio;
+                       nworkq = workq; /* starting pt */
+                       while (kernel_workq_count < KERNEL_WORKQ_ELEM_MAX) {
+                               headp->next_workq = TAILQ_NEXT(workq, wq_list);
+                               if (headp->next_workq == NULL)
+                                       headp->next_workq = TAILQ_FIRST(&headp->wqhead);
+#if WQ_TRACE
+       __kdebug_trace(0x9008098, kernel_workq_count, workq, 0,  1, 0);
+#endif
+                               val = post_nextworkitem(workq);
+
+                               if (val != 0) {
+                                       /* things could have changed so reasses */
+                                       /* If kernel queue is full , skip */
+                                       if (kernel_workq_count >= KERNEL_WORKQ_ELEM_MAX)
+                                               break;
+                                       /* If anything with higher prio arrived, then reevaluate */
+                                       if (wqreadyprio < curwqprio)
+                                               goto loop;      /* we need re evaluate again */
+                                       /* we can post some more work items */
+                                       found = 1;
+                               }
+
+                               /* cannot use workq here as it could be freed */
+                               if (TAILQ_EMPTY(&headp->wqhead))
+                                       break;
+                               /* if we found nothing to run and only one workqueue in the list, skip */
+                               if ((val == 0) && (workq == headp->next_workq))
+                                       break;
+                               workq = headp->next_workq;
+                               if (workq == NULL)
+                                       workq = TAILQ_FIRST(&headp->wqhead);
+                               if (val != 0)
+                                       nworkq = workq;
+                               /* if we found nothing to run and back to workq where we started */
+                               if ((val == 0) && (workq == nworkq))
+                                       break;
+                       }
+                       if (kernel_workq_count >= KERNEL_WORKQ_ELEM_MAX)
+                               break;
+               }
+               /* nothing found to run?  */
+               if (found == 0)
+                       break;
+       }
+       workqueue_list_unlock();
+}
+
+static int
+post_nextworkitem(pthread_workqueue_t workq)
+{
+       int error, prio;
+       pthread_workitem_t witem;
+       pthread_workqueue_head_t headp;
+       void (*func)(pthread_workqueue_t, void *);
+
+               if ((workq->flags & PTHREAD_WORKQ_SUSPEND) == PTHREAD_WORKQ_SUSPEND) {
+                       return(0);
+               }
+#if WQ_TRACE
+               __kdebug_trace(0x900809c, workq, workq->item_listhead.tqh_first, 0, 1, 0);
+#endif
+               if (TAILQ_EMPTY(&workq->item_listhead)) {
+                       return(0);
+               }
+               if ((workq->flags & PTHREAD_WORKQ_BARRIER_ON) ==  PTHREAD_WORKQ_BARRIER_ON)
+                               return(0);
+
+               witem = TAILQ_FIRST(&workq->item_listhead);
+               headp = workq->headp;
+#if WQ_TRACE
+               __kdebug_trace(0x900809c, workq, witem, 0, 0xee, 0);
+#endif
+               if ((witem->flags & PTH_WQITEM_BARRIER) == PTH_WQITEM_BARRIER) {
+#if WQ_TRACE
+                       __kdebug_trace(0x9000064, workq, 0, 0, 2, 0);
+#endif
+                       
+                       if ((witem->flags & PTH_WQITEM_APPLIED) != 0) {
+                               return(0);
+                       }
+                       /* Also barrier when nothing is there needs to be handled */
+                       /* Nothing to wait for */
+                       if (workq->kq_count != 0) {
+                               witem->flags |= PTH_WQITEM_APPLIED;
+                               workq->flags |= PTHREAD_WORKQ_BARRIER_ON;
+                               workq->barrier_count = workq->kq_count;
+#if WQ_TRACE
+                       __kdebug_trace(0x9000064, 1, workq->barrier_count, 0, 0, 0);
+#endif
+                               return(1);
+                       } else {
+#if WQ_TRACE
+                       __kdebug_trace(0x9000064, 2, workq->barrier_count, 0, 0, 0);
+#endif
+                               if (witem->func != NULL) {
+                               /* since we are going to drop list lock */
+                               witem->flags |= PTH_WQITEM_APPLIED;
+                               workq->flags |= PTHREAD_WORKQ_BARRIER_ON;
+                                       workqueue_list_unlock();
+                                       func = (void (*)(pthread_workqueue_t, void *))witem->func;
+                                       (*func)(workq, witem->func_arg);
+#if WQ_TRACE
+                       __kdebug_trace(0x9000064, 3, workq->barrier_count, 0, 0, 0);
+#endif
+                                       workqueue_list_lock();
+                                       workq->flags &= ~PTHREAD_WORKQ_BARRIER_ON;
+                               }
+                               TAILQ_REMOVE(&workq->item_listhead, witem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080a8, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+                               free_workitem(witem);
+#if WQ_TRACE
+                       __kdebug_trace(0x9000064, 4, workq->barrier_count, 0, 0, 0);
+#endif
+                               return(1);
+                       }
+               } else if ((witem->flags & PTH_WQITEM_DESTROY) == PTH_WQITEM_DESTROY) {
+#if WQ_TRACE
+                       __kdebug_trace(0x9000068, 1, workq->kq_count, 0, 0, 0);
+#endif
+                       if ((witem->flags & PTH_WQITEM_APPLIED) != 0) {
+                               return(0);
+                       }
+                       witem->flags |= PTH_WQITEM_APPLIED;
+                       workq->flags |= (PTHREAD_WORKQ_BARRIER_ON | PTHREAD_WORKQ_TERM_ON);
+                       workq->barrier_count = workq->kq_count;
+                       workq->term_callback = (void (*)(struct _pthread_workqueue *,void *))witem->func;
+                       workq->term_callarg = witem->func_arg;
+                       TAILQ_REMOVE(&workq->item_listhead, witem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080a8, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+                       if ((TAILQ_EMPTY(&workq->item_listhead)) && (workq->kq_count == 0)) {
+                               if (!(TAILQ_EMPTY(&workq->item_kernhead))) {
+#if WQ_TRACE
+                               __kdebug_trace(0x900006c, workq, workq->kq_count, 0, 0xff, 0);
+#endif
+                               }
+                               free_workitem(witem);
+                               workq->flags |= PTHREAD_WORKQ_DESTROYED;
+#if WQ_TRACE
+                       __kdebug_trace(0x900006c, workq, workq->kq_count, 0, 1, 0);
+#endif
+                               headp = __pthread_wq_head_tbl[workq->queueprio];
+                               if (headp->next_workq == workq) {
+                                       headp->next_workq = TAILQ_NEXT(workq, wq_list);
+                                       if (headp->next_workq == NULL) {
+                                               headp->next_workq = TAILQ_FIRST(&headp->wqhead);
+                                               if (headp->next_workq == workq)
+                                                       headp->next_workq = NULL;
+                                       }
+                               }
+                               workq->sig = 0;
+                               TAILQ_REMOVE(&headp->wqhead, workq, wq_list);
+                               if (workq->term_callback != NULL) {
+                                       workqueue_list_unlock();
+                                       (*workq->term_callback)(workq, workq->term_callarg);
+                                       workqueue_list_lock();
+                               }
+                               free_workqueue(workq);
+                               return(1);
+                       } else  {
+                               TAILQ_INSERT_HEAD(&workq->item_listhead, witem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080b0, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+                       }
+#if WQ_TRACE
+                       __kdebug_trace(0x9000068, 2, workq->barrier_count, 0, 0, 0);
+#endif
+                       return(1);
+               }  else {
+#if WQ_TRACE
+                       __kdebug_trace(0x9000060, witem, workq, witem->func_arg, 0xfff, 0);
+#endif
+                       TAILQ_REMOVE(&workq->item_listhead, witem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080a8, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+                       TAILQ_INSERT_TAIL(&workq->item_kernhead, witem, item_entry);
+                       if ((witem->flags & PTH_WQITEM_KERN_COUNT) == 0) {
+                               workq->kq_count++;
+                               witem->flags |= PTH_WQITEM_KERN_COUNT;
+                       }
+                       OSAtomicIncrement32(&kernel_workq_count);
+                       workqueue_list_unlock();
+                       
+                       prio = workq->queueprio;
+                       if (workq->overcommit != 0) {
+                               prio |= WORKQUEUE_OVERCOMMIT;
+                       }
+
+                       if (( error =__workq_kernreturn(WQOPS_QUEUE_ADD, witem, workq->affinity, prio)) == -1) {
+                               OSAtomicDecrement32(&kernel_workq_count);
+                               workqueue_list_lock();
+#if WQ_TRACE
+                       __kdebug_trace(0x900007c, witem, workq, witem->func_arg, workq->kq_count, 0);
+#endif
+                               TAILQ_REMOVE(&workq->item_kernhead, witem, item_entry);
+                               TAILQ_INSERT_HEAD(&workq->item_listhead, witem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080b0, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+                               if ((workq->flags & (PTHREAD_WORKQ_BARRIER_ON | PTHREAD_WORKQ_TERM_ON)) != 0)
+                                       workq->flags |= PTHREAD_WORKQ_REQUEUED;
+                       } else
+                               workqueue_list_lock();
+#if WQ_TRACE
+                       __kdebug_trace(0x9000060, witem, workq, witem->func_arg, workq->kq_count, 0);
+#endif
+                       return(1);
+               }
+               /* noone should come here */
+#if 1
+               printf("error in logic for next workitem\n");
+               LIBC_ABORT("error in logic for next workitem");
+#endif
+               return(0);
+}
+
+void
+_pthread_wqthread(pthread_t self, mach_port_t kport, void * stackaddr, pthread_workitem_t item, int reuse)
+{
+       int ret;
+       pthread_attr_t *attrs = &_pthread_attr_default;
+       pthread_workqueue_t workq;
+#if WQ_DEBUG
+       pthread_t pself;
+#endif
+
+
+       workq = item->workq;
+       if (reuse == 0) {
+               /* reuse is set to 0, when a thread is newly created to run a workitem */
+               _pthread_struct_init(self, attrs, stackaddr,  DEFAULT_STACK_SIZE, 1, 1);
+               self->wqthread = 1;
+               self->wqkillset = 0;
+               self->parentcheck = 1;
+
+               /* These are not joinable threads */
+               self->detached &= ~PTHREAD_CREATE_JOINABLE;
+               self->detached |= PTHREAD_CREATE_DETACHED;
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__) 
+               _pthread_set_self(self);
+#endif
+#if WQ_TRACE
+               __kdebug_trace(0x9000050, self, item, item->func_arg, 0, 0);
+#endif
+               self->kernel_thread = kport;
+               self->fun = (void *(*)(void *))item->func;
+               self->arg = item->func_arg;
+               /* Add to the pthread list */
+               LOCK(_pthread_list_lock);
+               TAILQ_INSERT_TAIL(&__pthread_head, self, plist);
+#if PTH_LISTTRACE
+               __kdebug_trace(0x900000c, self, 0, 0, 10, 0);
+#endif
+               _pthread_count++;
+               UNLOCK(_pthread_list_lock);
+
+#if defined(__i386__) || defined(__x86_64__) || defined(__arm__)
+               if( (self->thread_id = __thread_selfid()) == (__uint64_t)-1)
+                       printf("Failed to set thread_id in pthread_wqthread\n");
+#endif
+
+       } else  {
+               /* reuse is set to 1, when a thread is resued to run another work item */
+#if WQ_TRACE
+               __kdebug_trace(0x9000054, self, item, item->func_arg, 0, 0);
+#endif
+               /* reset all tsd from 1 to KEYS_MAX */
+               if (self == NULL)
+                       LIBC_ABORT("_pthread_wqthread: pthread %p setup to be NULL", self);
+
+               self->fun = (void *(*)(void *))item->func;
+               self->arg = item->func_arg;
+       }
+
+#if WQ_DEBUG
+       if (reuse == 0) {
+               pself = pthread_self();
+               if (self != pself) {
+#if WQ_TRACE
+               __kdebug_trace(0x9000078, self, pself, item->func_arg, 0, 0);
+#endif
+                       printf("pthread_self not set: pself %p, passed in %p\n", pself, self);
+                       _pthread_set_self(self);
+                       pself = pthread_self();
+                       if (self != pself)
+                               printf("(2)pthread_self not set: pself %p, passed in %p\n", pself, self);
+                       pself = self;
+               }
+       } else {
+               pself = pthread_self();
+               if (self != pself) {
+                       printf("(3)pthread_self not set in reuse: pself %p, passed in %p\n", pself, self);
+                       LIBC_ABORT("(3)pthread_self not set in reuse: pself %p, passed in %p", pself, self);
+               }
+       }
+#endif /* WQ_DEBUG */
+
+       self->cur_workq = workq;
+       self->cur_workitem = item;
+       OSAtomicDecrement32(&kernel_workq_count);
+
+       ret = (int)(intptr_t)(*self->fun)(self->arg);
+
+       /* If we reach here without going through the above initialization path then don't go through
+        * with the teardown code path ( e.g. setjmp/longjmp ). Instead just exit this thread.
+        */
+       if(self != pthread_self()) {
+               pthread_exit(PTHREAD_CANCELED);
+       }
+       
+       workqueue_exit(self, workq, item);
+
+}
+
+static void
+workqueue_exit(pthread_t self, pthread_workqueue_t workq, pthread_workitem_t item)
+{
+       pthread_workitem_t baritem;
+       pthread_workqueue_head_t headp;
+       void (*func)(pthread_workqueue_t, void *);
+       
+       workqueue_list_lock();
+
+       TAILQ_REMOVE(&workq->item_kernhead, item, item_entry);
+       workq->kq_count--;
+#if WQ_TRACE
+       __kdebug_trace(0x9000070, self, 1, item->func_arg, workq->kq_count, 0);
+#endif
+       free_workitem(item);
+
+       if ((workq->flags & PTHREAD_WORKQ_BARRIER_ON) == PTHREAD_WORKQ_BARRIER_ON) {
+               workq->barrier_count--;
+#if WQ_TRACE
+               __kdebug_trace(0x9000084, self, workq->barrier_count, workq->kq_count, 1, 0);
+#endif
+               if (workq->barrier_count <= 0 ) {
+                       /* Need to remove barrier item from the list */
+                       baritem = TAILQ_FIRST(&workq->item_listhead);
+#if WQ_DEBUG
+                       if ((baritem->flags & (PTH_WQITEM_BARRIER | PTH_WQITEM_DESTROY| PTH_WQITEM_APPLIED)) == 0)
+                               printf("Incorect bar item being removed in barrier processing\n");
+#endif /* WQ_DEBUG */
+                       /* if the front item is a barrier and call back is registered, run that */
+                        if (((baritem->flags & PTH_WQITEM_BARRIER) == PTH_WQITEM_BARRIER) && (baritem->func != NULL)) {
+                               workqueue_list_unlock();
+                               func = (void (*)(pthread_workqueue_t, void *))baritem->func;
+                               (*func)(workq, baritem->func_arg);
+                               workqueue_list_lock();
+                       }
+                       TAILQ_REMOVE(&workq->item_listhead, baritem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080a8, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+                       free_workitem(baritem);
+                       workq->flags &= ~PTHREAD_WORKQ_BARRIER_ON;
+#if WQ_TRACE
+                       __kdebug_trace(0x9000058, self, item, item->func_arg, 0, 0);
+#endif
+                       if ((workq->flags & PTHREAD_WORKQ_TERM_ON) != 0) {
+                               headp = __pthread_wq_head_tbl[workq->queueprio];
+                               workq->flags |= PTHREAD_WORKQ_DESTROYED;
+#if WQ_TRACE
+                       __kdebug_trace(0x900006c, workq, workq->kq_count, 0, 2, 0);
+#endif
+                               if (headp->next_workq == workq) {
+                                       headp->next_workq = TAILQ_NEXT(workq, wq_list);
+                                       if (headp->next_workq == NULL) {
+                                               headp->next_workq = TAILQ_FIRST(&headp->wqhead);
+                                               if (headp->next_workq == workq)
+                                                       headp->next_workq = NULL;
+                                       }
+                               }
+                               TAILQ_REMOVE(&headp->wqhead, workq, wq_list);
+                               workq->sig = 0;
+                               if (workq->term_callback != NULL) {
+                                       workqueue_list_unlock();
+                                       (*workq->term_callback)(workq, workq->term_callarg);
+                                       workqueue_list_lock();
+                               }
+                               free_workqueue(workq);
+                       } else {
+                       /* if there are higher prio schedulabel item reset to wqreadyprio */
+                               if ((workq->queueprio < wqreadyprio) && (!(TAILQ_EMPTY(&workq->item_listhead))))
+                                               wqreadyprio = workq->queueprio;
+                       }
+               }
+       } 
+#if WQ_TRACE
+       else {
+               __kdebug_trace(0x9000070, self, 2, item->func_arg, workq->barrier_count, 0);
+       }
+       
+       __kdebug_trace(0x900005c, self, item, 0, 0, 0);
+#endif
+       pick_nextworkqueue_droplock();
+       _pthread_workq_return(self);
+}
+
+static void 
+_pthread_workq_return(pthread_t self)
+{
+       __workq_kernreturn(WQOPS_THREAD_RETURN, NULL, 0, 0);
+
+       /* This is the way to terminate the thread */
+       _pthread_exit(self, NULL);
+}
+
+
+/* XXXXXXXXXXXXX Pthread Workqueue functions XXXXXXXXXXXXXXXXXX */
+
+int 
+pthread_workqueue_create_np(pthread_workqueue_t * workqp, const pthread_workqueue_attr_t * attr)
+{
+       pthread_workqueue_t wq;
+       pthread_workqueue_head_t headp;
+
+#if defined(__ppc__)
+       IF_ROSETTA() {
+               return(ENOTSUP);
+       }
+#endif
+       if ((attr != NULL) && (attr->sig != PTHREAD_WORKQUEUE_ATTR_SIG)) {
+               return(EINVAL);
+       }
+
+       if (__is_threaded == 0)
+               __is_threaded = 1;
+
+       workqueue_list_lock();
+       if (kernel_workq_setup == 0) {
+               int ret = _pthread_work_internal_init();
+               if (ret != 0) {
+                       workqueue_list_unlock();
+                       return(ret);
+               }
+       }
+       
+       wq = alloc_workqueue();
+       
+       _pthread_workq_init(wq, attr);
+
+       headp = __pthread_wq_head_tbl[wq->queueprio];
+       TAILQ_INSERT_TAIL(&headp->wqhead, wq, wq_list);
+       if (headp->next_workq == NULL) {
+               headp->next_workq = TAILQ_FIRST(&headp->wqhead);
+       }
+
+       workqueue_list_unlock();
+       
+       *workqp = wq;
+
+       return(0);
+}
+
+int 
+pthread_workqueue_additem_np(pthread_workqueue_t workq, void ( *workitem_func)(void *), void * workitem_arg, pthread_workitem_handle_t * itemhandlep, unsigned int *gencountp)
+{
+       pthread_workitem_t witem;
+
+       if (valid_workq(workq) == 0) {
+               return(EINVAL);
+       }
+
+       workqueue_list_lock();
+
+       /*
+        * Allocate the workitem here as it can drop the lock.
+        * Also we can evaluate the workqueue state only once.
+        */
+       witem = alloc_workitem();
+       witem->func = workitem_func;
+       witem->func_arg = workitem_arg;
+       witem->workq = workq;
+
+       /* alloc workitem can drop the lock, check the state  */
+       if ((workq->flags & (PTHREAD_WORKQ_IN_TERMINATE | PTHREAD_WORKQ_DESTROYED)) != 0) {
+               free_workitem(witem);
+               workqueue_list_unlock();
+               *itemhandlep = 0;
+               return(ESRCH);
+       }
+
+       if (itemhandlep != NULL)
+               *itemhandlep = (pthread_workitem_handle_t *)witem;
+       if (gencountp != NULL)
+               *gencountp = 0;
+#if WQ_TRACE
+       __kdebug_trace(0x9008090, witem, witem->func, witem->func_arg,  workq, 0);
+#endif
+       TAILQ_INSERT_TAIL(&workq->item_listhead, witem, item_entry);
+#if WQ_LISTTRACE
+       __kdebug_trace(0x90080a4, workq, &workq->item_listhead, workq->item_listhead.tqh_first,  workq->item_listhead.tqh_last, 0);
+#endif
+       
+       if (((workq->flags & PTHREAD_WORKQ_BARRIER_ON) == 0) && (workq->queueprio < wqreadyprio))
+                       wqreadyprio = workq->queueprio;
+
+       pick_nextworkqueue_droplock();
+
+       return(0);
+}
+
+int
+pthread_workqueue_getovercommit_np(pthread_workqueue_t workq,  unsigned int *ocommp)
+{
+        if (valid_workq(workq) == 0) {
+                return(EINVAL);
+        }
+
+       if (ocommp != NULL)
+               *ocommp = workq->overcommit;
+       return(0);
+}
+
+
+#else /* !BUILDING_VARIANT ] [ */
+extern int __unix_conforming;
+extern int _pthread_count;
+extern pthread_lock_t _pthread_list_lock;
+extern void _pthread_testcancel(pthread_t thread, int isconforming);
+extern int _pthread_reap_thread(pthread_t th, mach_port_t kernel_thread, void **value_ptr, int conforming);
+
+#endif /* !BUILDING_VARIANT ] */
+
+#if __DARWIN_UNIX03
+
+__private_extern__ void 
+__posix_join_cleanup(void *arg)
+{
+       pthread_t thread = (pthread_t)arg;
+       int already_exited, res;
+       void * dummy;
+       semaphore_t death;
+       int newstyle;
+
+       LOCK(thread->lock);
+       already_exited = (thread->detached & _PTHREAD_EXITED);
+
+       newstyle = thread->newstyle;
+       
+#if WQ_TRACE
+       __kdebug_trace(0x900002c, thread, newstyle, 0, 0, 0);
+#endif
+       if (newstyle == 0) {
+               death = thread->death;
+               if (!already_exited){
+                       thread->joiner = (struct _pthread *)NULL;
+                       UNLOCK(thread->lock);
+                       restore_sem_to_pool(death);
+               } else {
+                       UNLOCK(thread->lock);
+                       while ((res = _pthread_reap_thread(thread,
+                                       thread->kernel_thread,
+                                       &dummy, 1)) == EAGAIN)
+                               {
+                                       sched_yield();
+                               }
+                       restore_sem_to_pool(death);
+
+               }
+
+       } else {
+               /* leave another thread to join */
+               thread->joiner = (struct _pthread *)NULL;
+               UNLOCK(thread->lock);
+       }
+}
+
+#endif /* __DARWIN_UNIX03 */
+
+
+/*
+ * Wait for a thread to terminate and obtain its exit value.
+ */
+/*
+int       
+pthread_join(pthread_t thread, 
+            void **value_ptr)
+
+moved to pthread_cancelable.c */
+
+/*
+ * Cancel a thread
+ */
+int
+pthread_cancel(pthread_t thread)
+{
+#if __DARWIN_UNIX03
+       if (__unix_conforming == 0)
+               __unix_conforming = 1;
+#endif /* __DARWIN_UNIX03 */
+
+       if (_pthread_lookup_thread(thread, NULL, 0) != 0)
+               return(ESRCH);
+
+       /* if the thread is a workqueue thread, then return error */
+       if (thread->wqthread != 0) {
+               return(ENOTSUP);
+       }
+#if __DARWIN_UNIX03
+       int state;
+
+       LOCK(thread->lock);
+       state = thread->cancel_state |= _PTHREAD_CANCEL_PENDING;
+       UNLOCK(thread->lock);
+       if (state & PTHREAD_CANCEL_ENABLE)
+               __pthread_markcancel(thread->kernel_thread);
+#else /* __DARWIN_UNIX03 */
+       thread->cancel_state |= _PTHREAD_CANCEL_PENDING;
+#endif /* __DARWIN_UNIX03 */
+       return (0);
+}
+
+void
+pthread_testcancel(void)
+{
+       pthread_t self = pthread_self();
+
+#if __DARWIN_UNIX03
+       if (__unix_conforming == 0)
+               __unix_conforming = 1;
+       _pthread_testcancel(self, 1);
+#else /* __DARWIN_UNIX03 */
+       _pthread_testcancel(self, 0);
+#endif /* __DARWIN_UNIX03 */
+
+}
+
+
+/*
+ * Query/update the cancelability 'state' of a thread
+ */
+int
+pthread_setcancelstate(int state, int *oldstate)
+{
+#if __DARWIN_UNIX03
+       if (__unix_conforming == 0) {
+               __unix_conforming = 1;
+       }
+       return (_pthread_setcancelstate_internal(state, oldstate, 1));
+#else /* __DARWIN_UNIX03 */
+       return (_pthread_setcancelstate_internal(state, oldstate, 0));
+#endif /* __DARWIN_UNIX03 */
+
+}
+
+
+
+/*
+ * Query/update the cancelability 'type' of a thread
+ */
+int
+pthread_setcanceltype(int type, int *oldtype)
+{
+       pthread_t self = pthread_self();
+       
+#if __DARWIN_UNIX03
+       if (__unix_conforming == 0)
+               __unix_conforming = 1;
+#endif /* __DARWIN_UNIX03 */
+
+       if ((type != PTHREAD_CANCEL_DEFERRED) && 
+           (type != PTHREAD_CANCEL_ASYNCHRONOUS))
+               return EINVAL;
+       self = pthread_self();
+       LOCK(self->lock);
+       if (oldtype)
+               *oldtype = self->cancel_state & _PTHREAD_CANCEL_TYPE_MASK;
+       self->cancel_state &= ~_PTHREAD_CANCEL_TYPE_MASK;
+       self->cancel_state |= type;
+       UNLOCK(self->lock);
+#if !__DARWIN_UNIX03
+       _pthread_testcancel(self, 0);  /* See if we need to 'die' now... */
+#endif /* __DARWIN_UNIX03 */
+       return (0);
+}
+
+int 
+pthread_sigmask(int how, const sigset_t * set, sigset_t * oset)
+{
+#if __DARWIN_UNIX03
+       int err = 0;
+
+       if (__pthread_sigmask(how, set, oset) == -1) {
+               err = errno;
+       } 
+       return(err);
+#else /* __DARWIN_UNIX03 */
+       return(__pthread_sigmask(how, set, oset));
+#endif /* __DARWIN_UNIX03 */
+}
+
+/*
+int
+sigwait(const sigset_t * set, int * sig)
+
+moved to pthread_cancelable.c */