[apple/libc.git] / pthreads / pthread.c

/*
 * Copyright 1996 1995 by Open Software Foundation, Inc. 1997 1996 1995 1994 1993 1992 1991  
 *              All Rights Reserved 
 *  
 * Permission to use, copy, modify, and distribute this software and 
 * its documentation for any purpose and without fee is hereby granted, 
 * provided that the above copyright notice appears in all copies and 
 * that both the copyright notice and this permission notice appear in 
 * supporting documentation. 
 *  
 * OSF DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE 
 * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 
 * FOR A PARTICULAR PURPOSE. 
 *  
 * IN NO EVENT SHALL OSF BE LIABLE FOR ANY SPECIAL, INDIRECT, OR 
 * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM 
 * LOSS OF USE, DATA OR PROFITS, WHETHER IN ACTION OF CONTRACT, 
 * NEGLIGENCE, OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION 
 * WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 
 * 
 */
/*
 * MkLinux
 */

/*
 * POSIX Pthread Library
 */

#define __POSIX_LIB__
#include <assert.h>
#include <stdio.h>      /* For printf(). */
#include <stdlib.h>
#include <errno.h>      /* For __mach_errno_addr() prototype. */
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/sysctl.h>
#include <sys/syscall.h>
#include <machine/vmparam.h>
#include <mach/vm_statistics.h>
#define __APPLE_API_PRIVATE
#include <machine/cpu_capabilities.h>

#include "pthread_internals.h"

/* Per-thread kernel support */
extern void _pthread_set_self(pthread_t);
extern void mig_init(int);

/* Get CPU capabilities from the kernel */
__private_extern__ void _init_cpu_capabilities(void);

/* Needed to tell the malloc subsystem we're going multithreaded */
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;

/* 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 */
static struct _pthread _thread = {0};

/* This global should be used (carefully) by anyone needing to know if a 
** pthread has been created.
*/
int __is_threaded = 0;
static int _pthread_count = 1;

static pthread_lock_t _pthread_count_lock = LOCK_INITIALIZER;

/* Same implementation as LOCK, but without the __is_threaded check */
int _spin_tries = 0;
__private_extern__ void _spin_lock_retry(pthread_lock_t *lock)
{ 
        int tries = _spin_tries;
        do {
                if (tries-- > 0)
                        continue;
                syscall_thread_switch(THREAD_NULL, SWITCH_OPTION_DEPRESS, 1);
                tries = _spin_tries;
        } while(!_spin_lock_try(lock));
}

extern mach_port_t thread_recycle_port;

/* These are used to keep track of a semaphore pool shared by mutexes and condition
** variables.
*/

static semaphore_t *sem_pool = NULL;
static int sem_pool_count = 0;
static int sem_pool_current = 0;
static pthread_lock_t sem_pool_lock = LOCK_INITIALIZER;

static int default_priority;
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)

#endif

#if defined(__ppc__)
static const vm_address_t PTHREAD_STACK_HINT = 0xF0000000;
#elif defined(__i386__)
static const vm_address_t PTHREAD_STACK_HINT = 0xB0000000;
#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 */

static int
_pthread_allocate_stack(pthread_attr_t *attrs, void **stack)
{
    kern_return_t kr;
#if 1
    assert(attrs->stacksize >= PTHREAD_STACK_MIN);
    if (attrs->stackaddr != NULL) {
        assert(((vm_address_t)(attrs->stackaddr) & (vm_page_size - 1)) == 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,
                        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,
                                        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

#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)
#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;
                        }
#else
                        if (kr == KERN_SUCCESS)
                            break;
#endif
                }
                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 */
                        }
#endif
                        free_stacks = (vm_address_t *)lowest_stack;
                        lowest_stack = 0;
                }
                *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;
}

static pthread_attr_t _pthread_attr_default = {0};

/*
 * Destroy a thread attribute structure
 */
int       
pthread_attr_destroy(pthread_attr_t *attr)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                return (ESUCCESS);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Get the 'detach' state from a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_getdetachstate(const pthread_attr_t *attr, 
                            int *detachstate)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *detachstate = attr->detached;
                return (ESUCCESS);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Get the 'inherit scheduling' info from a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_getinheritsched(const pthread_attr_t *attr, 
                             int *inheritsched)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *inheritsched = attr->inherit;
                return (ESUCCESS);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Get the scheduling parameters from a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_getschedparam(const pthread_attr_t *attr, 
                           struct sched_param *param)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *param = attr->param;
                return (ESUCCESS);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Get the scheduling policy from a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_getschedpolicy(const pthread_attr_t *attr, 
                            int *policy)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                *policy = attr->policy;
                return (ESUCCESS);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

static const size_t DEFAULT_STACK_SIZE = DFLSSIZ;
/*
 * 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->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);
}

/*
 * Set the 'detach' state in a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_setdetachstate(pthread_attr_t *attr, 
                            int detachstate)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                if ((detachstate == PTHREAD_CREATE_JOINABLE) ||
                    (detachstate == PTHREAD_CREATE_DETACHED))
                {
                        attr->detached = detachstate;
                        return (ESUCCESS);
                } else
                {
                        return (EINVAL);
                }
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Set the 'inherit scheduling' state in a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_setinheritsched(pthread_attr_t *attr, 
                             int inheritsched)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                if ((inheritsched == PTHREAD_INHERIT_SCHED) ||
                    (inheritsched == PTHREAD_EXPLICIT_SCHED))
                {
                        attr->inherit = inheritsched;
                        return (ESUCCESS);
                } else
                {
                        return (EINVAL);
                }
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Set the scheduling paramters in a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_setschedparam(pthread_attr_t *attr, 
                           const struct sched_param *param)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                /* TODO: Validate sched_param fields */
                attr->param = *param;
                return (ESUCCESS);
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Set the scheduling policy in a thread attribute structure.
 * Note: written as a helper function for info hiding
 */
int       
pthread_attr_setschedpolicy(pthread_attr_t *attr, 
                            int policy)
{
        if (attr->sig == _PTHREAD_ATTR_SIG)
        {
                if ((policy == SCHED_OTHER) ||
                    (policy == SCHED_RR) ||
                    (policy == SCHED_FIFO))
                {
                        attr->policy = policy;
                        return (ESUCCESS);
                } else
                {
                        return (EINVAL);
                }
        } else
        {
                return (EINVAL); /* Not an attribute structure! */
        }
}

/*
 * Set the scope for the thread.
 * We currently only provide PTHREAD_SCOPE_SYSTEM
 */
int
pthread_attr_setscope(pthread_attr_t *attr,
                            int scope)
{
    if (attr->sig == _PTHREAD_ATTR_SIG) {
        if (scope == PTHREAD_SCOPE_SYSTEM) {
            /* No attribute yet for the scope */
            return (ESUCCESS);
        } else if (scope == PTHREAD_SCOPE_PROCESS) {
            return (ENOTSUP);
        }
    }
    return (EINVAL); /* Not an attribute structure! */
}

/*
 * Get the scope for the thread.
 * We currently only provide PTHREAD_SCOPE_SYSTEM
 */
int
pthread_attr_getscope(pthread_attr_t *attr,
                            int *scope)
{
    if (attr->sig == _PTHREAD_ATTR_SIG) {
        *scope = PTHREAD_SCOPE_SYSTEM;
        return (ESUCCESS);
    }
    return (EINVAL); /* Not an attribute structure! */
}

/* Get the base stack address of the given thread */
int
pthread_attr_getstackaddr(const pthread_attr_t *attr, void **stackaddr)
{
    if (attr->sig == _PTHREAD_ATTR_SIG) {
        *stackaddr = attr->stackaddr;
        return (ESUCCESS);
    } else {
        return (EINVAL); /* Not an attribute structure! */
    }
}

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)) {
        attr->stackaddr = stackaddr;
        attr->freeStackOnExit = FALSE;
        return (ESUCCESS);
    } else {
        return (EINVAL); /* Not an attribute structure! */
    }
}

int
pthread_attr_getstacksize(const pthread_attr_t *attr, size_t *stacksize)
{
    if (attr->sig == _PTHREAD_ATTR_SIG) {
        *stacksize = attr->stacksize;
        return (ESUCCESS);
    } else {
        return (EINVAL); /* Not an attribute structure! */
    }
}

int
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);
    } else {
        return (EINVAL); /* Not an attribute structure! */
    }
}

int
pthread_attr_getstack(const pthread_attr_t *attr, void **stackaddr, size_t * stacksize)
{
    if (attr->sig == _PTHREAD_ATTR_SIG) {
        *stackaddr = attr->stackaddr;
        *stacksize = attr->stacksize;
        return (ESUCCESS);
    } else {
        return (EINVAL); /* Not an attribute structure! */
    }
}

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;
                attr->stacksize = stacksize;
                return (ESUCCESS);
    } else {
        return (EINVAL); /* Not an attribute structure! */
    }
}

/*
 * Create and start execution of a new thread.
 */

static void
_pthread_body(pthread_t self)
{
    _pthread_set_self(self);
    pthread_exit((self->fun)(self->arg));
}

int
_pthread_create(pthread_t t,
                const pthread_attr_t *attrs,
                void *stack,
                const mach_port_t kernel_thread)
{
        int res;
        res = ESUCCESS;
        
        do
        {
                memset(t, 0, sizeof(*t));
                t->stacksize = attrs->stacksize;
                t->stackaddr = (void *)stack;
                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->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 _pthread_handler_rec *)NULL;
                t->death = SEMAPHORE_NULL;

                if (kernel_thread != MACH_PORT_NULL)
                        pthread_setschedparam(t, t->policy, &t->param);
        } while (0);
        return (res);
}

/* Need to deprecate this in future */
int
_pthread_is_threaded(void)
{
    return __is_threaded;
}

/* Non portable public api to know whether this process has(had) atleast one thread 
 * apart from main thread. There could be race if there is a thread in the process of
 * creation at the time of call . It does not tell whether there are more than one thread
 * at this point of time.
 */
int
pthread_is_threaded_np(void)
{
    return (__is_threaded);
}

mach_port_t
pthread_mach_thread_np(pthread_t t)
{
    thread_t kernel_thread;

    /* Wait for the creator to initialize it */
    while ((kernel_thread = t->kernel_thread) == MACH_PORT_NULL)
        sched_yield();

    return kernel_thread;
}

size_t
pthread_get_stacksize_np(pthread_t t)
{
    return t->stacksize;
}

void *
pthread_get_stackaddr_np(pthread_t t)
{
    return t->stackaddr;
}

mach_port_t
_pthread_reply_port(pthread_t t)
{
    return t->reply_port;
}


/* returns non-zero if the current thread is the main thread */
int
pthread_main_np(void)
{
        pthread_t self = pthread_self();

        return ((self->detached & _PTHREAD_CREATE_PARENT) == _PTHREAD_CREATE_PARENT);
}

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 = ESUCCESS;

        /* 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;
                LOCK(_pthread_count_lock);
                _pthread_count++;
                UNLOCK(_pthread_count_lock);

                /* Send it on it's way */
                t->arg = arg;
                t->fun = start_routine;
                /* Now set it up to execute */
                _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);
}

int
pthread_create_suspended_np(pthread_t *thread,
           const pthread_attr_t *attr,
           void *(*start_routine)(void *),
           void *arg)
{
    return _pthread_create_suspended(thread, attr, start_routine, arg, 1);
}

/*
 * Make a thread 'undetached' - no longer 'joinable' with other threads.
 */
int       
pthread_detach(pthread_t thread)
{
        if (thread->sig == _PTHREAD_SIG)
        {
                LOCK(thread->lock);
                if (thread->detached & PTHREAD_CREATE_JOINABLE)
                {
                        if (thread->detached & _PTHREAD_EXITED) {
                                UNLOCK(thread->lock);
                                pthread_join(thread, NULL);
                                return ESUCCESS;
                        } else {
                                semaphore_t death = thread->death;

                                thread->detached &= ~PTHREAD_CREATE_JOINABLE;
                                thread->detached |= PTHREAD_CREATE_DETACHED;
                                UNLOCK(thread->lock);
                                if (death)
                                        (void) semaphore_signal(death);
                                return (ESUCCESS);
                        }
                } else {
                        UNLOCK(thread->lock);
                        return (EINVAL);
                }
        } else {
                return (ESRCH); /* Not a valid thread */
        }
}


/* 
 * pthread_kill call to system call
 */


int   
pthread_kill (
        pthread_t th,
        int sig)
{
        int error = 0;
        
        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);
        }
        else
                return(ESRCH);
}

/* Announce that there are pthread resources ready to be reclaimed in a */
/* subsequent pthread_exit or reaped by pthread_join. In either case, the Mach */
/* 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;
        kern_return_t ret;

        msg.header.msgh_bits = MACH_MSGH_BITS(MACH_MSG_TYPE_MAKE_SEND,
                                              MACH_MSG_TYPE_MOVE_SEND);
        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;
        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) {
        mach_port_type_t ptype;
        kern_return_t ret;
        task_t self;

        self = mach_task_self();
        if (kernel_thread != MACH_PORT_DEAD) {
                ret = mach_port_type(self, kernel_thread, &ptype);
                if (ret == KERN_SUCCESS && ptype != MACH_PORT_TYPE_DEAD_NAME) {
                        /* not quite dead yet... */
                        return EAGAIN;
                }
                ret = mach_port_deallocate(self, kernel_thread);
                if (ret != KERN_SUCCESS) {
                        fprintf(stderr,
                                  "mach_port_deallocate(kernel_thread) failed: %s\n",
                                  mach_error_string(ret));
                }
        }

        if (th->reply_port != MACH_PORT_NULL) {
                ret = mach_port_mod_refs(self, th->reply_port,
                                                 MACH_PORT_RIGHT_RECEIVE, -1);
                if (ret != KERN_SUCCESS) {
                        fprintf(stderr,
                                  "mach_port_mod_refs(reply_port) failed: %s\n",
                                  mach_error_string(ret));
                }
        }

        if (th->freeStackOnExit) {
                vm_address_t addr = (vm_address_t)th->stackaddr;
                vm_size_t size;

                size = (vm_size_t)th->stacksize + vm_page_size;
                
#if !defined(STACK_GROWS_UP)
                addr -= size;
#endif
                ret = vm_deallocate(self, addr, size);
                if (ret != KERN_SUCCESS) {
                  fprintf(stderr,
                                    "vm_deallocate(stack) failed: %s\n",
                                    mach_error_string(ret));
                }
        }

        if (value_ptr)
                *value_ptr = th->exit_value;

        if (th != &_thread)
                free(th);

        return ESUCCESS;
}

static
void _pthread_reap_threads(void)
{
        mach_msg_empty_rcv_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, 
                        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;

                if (_pthread_reap_thread(thread, kernel_thread, (void **)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, 
                                MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
        }
}

/* For compatibility... */

pthread_t
_pthread_self() {
        return pthread_self();
}

/*
 * Terminate a thread.
 */
void 
pthread_exit(void *value_ptr)
{
        struct _pthread_handler_rec *handler;
        pthread_t self = pthread_self();
        kern_return_t kern_res;
        int thread_count;

        /* Make this thread not to receive any signals */
        syscall(331,1);

        while ((handler = self->cleanup_stack) != 0)
        {
                (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));
                }
        } 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);

        /* 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);

                                if (!already_exited)
                                {
                                        /* Wait for it to signal... */ 
                                        do {
                                                PTHREAD_MACH_CALL(semaphore_wait(death), kern_res);
                                        } while (kern_res != KERN_SUCCESS);
                                }

                                /* ... 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;
                        }
                } else {
                        UNLOCK(thread->lock);
                        res = EINVAL;
                }
                restore_sem_to_pool(death);
                return res;
        }
        return ESRCH;
}

/*
 * Get the scheduling policy and scheduling paramters for a thread.
 */
int       
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 */
        }
}

/*
 * Set the scheduling policy and scheduling paramters for a thread.
 */
int       
pthread_setschedparam(pthread_t thread, 
                      int policy,
                      const struct sched_param *param)
{
        policy_base_data_t bases;
        policy_base_t base;
        mach_msg_type_number_t count;
        kern_return_t ret;

        if (thread->sig == _PTHREAD_SIG)
        {
                switch (policy)
                {
                case SCHED_OTHER:
                        bases.ts.base_priority = param->sched_priority;
                        base = (policy_base_t)&bases.ts;
                        count = POLICY_TIMESHARE_BASE_COUNT;
                        break;
                case SCHED_FIFO:
                        bases.fifo.base_priority = param->sched_priority;
                        base = (policy_base_t)&bases.fifo;
                        count = POLICY_FIFO_BASE_COUNT;
                        break;
                case SCHED_RR:
                        bases.rr.base_priority = param->sched_priority;
                        /* quantum isn't public yet */
                        bases.rr.quantum = param->quantum;
                        base = (policy_base_t)&bases.rr;
                        count = POLICY_RR_BASE_COUNT;
                        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)
                {
                        return (EINVAL);
                }
                return (ESUCCESS);
        } else
        {
                return (ESRCH);  /* Not a valid thread structure */
        }
}

/*
 * Get the minimum priority for the given policy
 */
int
sched_get_priority_min(int policy)
{
    return default_priority - 16;
}

/*
 * Get the maximum priority for the given policy
 */
int
sched_get_priority_max(int policy)
{
    return default_priority + 16;
}

/*
 * Determine if two thread identifiers represent the same thread.
 */
int       
pthread_equal(pthread_t t1, 
              pthread_t t2)
{
        return (t1 == t2);
}

void 
cthread_set_self(void *cself)
{
    pthread_t self = pthread_self();
    if ((self == (pthread_t)NULL) || (self->sig != _PTHREAD_SIG)) {
        _pthread_set_self(cself);
        return;
    }
    self->cthread_self = cself;
}

void *
ur_cthread_self(void) {
    pthread_t self = pthread_self();
    if ((self == (pthread_t)NULL) || (self->sig != _PTHREAD_SIG)) {
        return (void *)self;
    }
    return self->cthread_self;
}

/*
 * Execute a function exactly one time in a thread-safe fashion.
 */
int       
pthread_once(pthread_once_t *once_control, 
             void (*init_routine)(void))
{
        LOCK(once_control->lock);
        if (once_control->sig == _PTHREAD_ONCE_SIG_init)
        {
                (*init_routine)();
                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);
        }
}

/*
 * Insert a cancellation point in a thread.
 */
static void
_pthread_testcancel(pthread_t thread)
{
        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);
        }
        UNLOCK(thread->lock);
}

void
pthread_testcancel(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);
}

int
pthread_setconcurrency(int new_level)
{
        pthread_concurrency = new_level;
        return(ESUCCESS);
}

/*
 * Perform package initialization - called automatically when application starts
 */

static 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];
        size_t len;
        int numcpus;

        count = HOST_PRIORITY_INFO_COUNT;
        info = (host_info_t)&priority_info;
        flavor = HOST_PRIORITY_INFO;
        host = mach_host_self();
        kr = host_info(host, flavor, info, &count);
        if (kr != KERN_SUCCESS)
                printf("host_info failed (%d); probably need privilege.\n", kr);
        else {
                default_priority = priority_info.user_priority;
                min_priority = priority_info.minimum_priority;
                max_priority = priority_info.maximum_priority;
        }
        attrs = &_pthread_attr_default;
        pthread_attr_init(attrs);

        thread = &_thread;
        _pthread_set_self(thread);
        _pthread_create(thread, attrs, (void *)USRSTACK, mach_thread_self());
        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;
                }
        }

        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);
    }
#endif
    
        mig_init(1);            /* enable multi-threaded mig interfaces */
        return 0;
}

int sched_yield(void)
{
    swtch_pri(0);
    return 0;
}

/* This is the "magic" that gets the initialization routine called when the application starts */
int (*_cthread_init_routine)(void) = pthread_init;

/* Get a semaphore from the pool, growing it if necessary */

__private_extern__ semaphore_t new_sem_from_pool(void) {
        kern_return_t res;
        semaphore_t sem;
        int i;
        
        LOCK(sem_pool_lock);
        if (sem_pool_current == sem_pool_count) {
                sem_pool_count += 16;
                sem_pool = realloc(sem_pool, sem_pool_count * sizeof(semaphore_t));
                for (i = sem_pool_current; i < sem_pool_count; i++) {
                        PTHREAD_MACH_CALL(semaphore_create(mach_task_self(), &sem_pool[i], SYNC_POLICY_FIFO, 0), res);
                }
        }
        sem = sem_pool[sem_pool_current++];
        UNLOCK(sem_pool_lock);
        return sem;
}

/* Put a semaphore back into the pool */
__private_extern__ void restore_sem_to_pool(semaphore_t sem) {
        LOCK(sem_pool_lock);
        sem_pool[--sem_pool_current] = sem;
        UNLOCK(sem_pool_lock);
}

static void sem_pool_reset(void) {
        LOCK(sem_pool_lock);
        sem_pool_count = 0;
        sem_pool_current = 0;
        sem_pool = NULL; 
        UNLOCK(sem_pool_lock);
}

__private_extern__ void _pthread_fork_child(void) {
        /* Just in case somebody had it locked... */
        UNLOCK(sem_pool_lock);
        sem_pool_reset();
        UNLOCK(_pthread_count_lock);
        _pthread_count = 1;
}