[apple/libc.git] / gen / nanosleep.c

/*
 * Copyright (c) 1999, 2003, 2006, 2007 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@
 */

#include <errno.h>
#include <sys/time.h>
#include <mach/mach_error.h>
#include <mach/mach_time.h>
#include <stdio.h>


#if __DARWIN_UNIX03
#include "pthread_internals.h"
#include <mach/clock.h>

extern int __unix_conforming;
extern mach_port_t clock_port;
extern semaphore_t clock_sem;
#ifdef VARIANT_CANCELABLE
extern void _pthread_testcancel(pthread_t thread, int isconforming);
extern int __semwait_signal(int cond_sem, int mutex_sem, int timeout, int relative, time_t tv_sec, __int32_t tv_nsec);
#define	SEMWAIT_SIGNAL	__semwait_signal
#else /* !VARIANT_CANCELABLE */
extern int __semwait_signal_nocancel(int cond_sem, int mutex_sem, int timeout, int relative, time_t tv_sec, __int32_t tv_nsec);
#define	SEMWAIT_SIGNAL	__semwait_signal_nocancel
#endif /* VARIANT_CANCELABLE */

int
nanosleep(const struct timespec *requested_time, struct timespec *remaining_time) {
    kern_return_t kret;
    int ret;
    mach_timespec_t remain;
    mach_timespec_t current;
   
	if (__unix_conforming == 0)
		__unix_conforming = 1;

#ifdef VARIANT_CANCELABLE
    _pthread_testcancel(pthread_self(), 1);
#endif /* VARIANT_CANCELABLE */

    if ((requested_time == NULL) || (requested_time->tv_sec < 0) || (requested_time->tv_nsec >= NSEC_PER_SEC)) {
        errno = EINVAL;
        return -1;
    }


    if (remaining_time != NULL) {
        kret = clock_get_time(clock_port, &current);
        if (kret != KERN_SUCCESS) {
            fprintf(stderr, "clock_get_time() failed: %s\n", mach_error_string(kret));
            return -1;
        }
    }
    ret = SEMWAIT_SIGNAL(clock_sem, MACH_PORT_NULL, 1, 1, requested_time->tv_sec, requested_time->tv_nsec);
    if (ret < 0) {
        if (errno == ETIMEDOUT) {
		return 0;
        } else if (errno == EINTR) {
            if (remaining_time != NULL) {
                ret = clock_get_time(clock_port, &remain);
                if (ret != KERN_SUCCESS) {
                    fprintf(stderr, "clock_get_time() failed: %s\n", mach_error_string(ret));
                    return -1;
                }
                /* This depends on the layout of a mach_timespec_t and timespec_t being equivalent */
                ADD_MACH_TIMESPEC(&current, requested_time);
                SUB_MACH_TIMESPEC(&current, &remain);
                remaining_time->tv_sec = current.tv_sec;
                remaining_time->tv_nsec = current.tv_nsec;
            }
        } else {
            errno = EINVAL;
	}
    }
    return -1;
}


#else /* !__DARWIN_UNIX03 */

typedef struct {
    uint64_t high;
    uint64_t low;
} uint128_t;

/* 128-bit addition: acc += add */
static inline void
add128_128(uint128_t *acc, uint128_t *add)
{
    acc->high += add->high;
    acc->low += add->low;
    if(acc->low < add->low)
	acc->high++; // carry
}

/* 128-bit subtraction: acc -= sub */
static inline void
sub128_128(uint128_t *acc, uint128_t *sub)
{
    acc->high -= sub->high;
    if(acc->low < sub->low)
	acc->high--; // borrow
    acc->low -= sub->low;
}

#define TWO64	(((double)(1ULL << 32)) * ((double)(1ULL << 32)))

static inline double
uint128_double(uint128_t *u)
{
    return TWO64 * u->high + u->low; // may loses precision
}

/* 64x64 -> 128 bit multiplication */
static inline void
mul64x64(uint64_t x, uint64_t y, uint128_t *prod)
{
    uint128_t add;
    /*
     * Split the two 64-bit multiplicands into 32-bit parts:
     * x => 2^32 * x1 + x2
     * y => 2^32 * y1 + y2
     */
    uint32_t x1 = (uint32_t)(x >> 32);
    uint32_t x2 = (uint32_t)x;
    uint32_t y1 = (uint32_t)(y >> 32);
    uint32_t y2 = (uint32_t)y;
    /*
     * direct multiplication:
     * x * y => 2^64 * (x1 * y1) + 2^32 (x1 * y2 + x2 * y1) + (x2 * y2)
     * The first and last terms are direct assignmenet into the uint128_t
     * structure.  Then we add the middle two terms separately, to avoid
     * 64-bit overflow.  (We could use the Karatsuba algorithm to save
     * one multiply, but it is harder to deal with 64-bit overflows.)
     */
    prod->high = (uint64_t)x1 * (uint64_t)y1;
    prod->low = (uint64_t)x2 * (uint64_t)y2;
    add.low = (uint64_t)x1 * (uint64_t)y2;
    add.high = (add.low >> 32);
    add.low <<= 32;
    add128_128(prod, &add);
    add.low = (uint64_t)x2 * (uint64_t)y1;
    add.high = (add.low >> 32);
    add.low <<= 32;
    add128_128(prod, &add);
}

/* calculate (x * y / divisor), using 128-bit internal calculations */
static int
muldiv128(uint64_t x, uint64_t y, uint64_t divisor, uint64_t *res)
{
    uint128_t temp;
    uint128_t divisor128 = {0, divisor};
    uint64_t result = 0;
    double recip;

    /* calculate (x * y) */
    mul64x64(x, y, &temp);
    /*
     * Now divide by the divisor.  We use floating point to calculate an
     * approximate answer and update the results.  Then we iterate and
     * calculate a correction from the difference.
     */
    recip = 1.0 / ((double)divisor);
    while(temp.high || temp.low >= divisor) {
	uint128_t backmul;
	uint64_t uapprox;
	double approx = uint128_double(&temp) * recip;

	if(approx > __LONG_LONG_MAX__)
	    return 0; // answer overflows 64-bits
	uapprox = (uint64_t)approx;
	mul64x64(uapprox, divisor, &backmul);
	/*
	 * Because we are using unsigned integers, we need to approach the
	 * answer from the lesser side.  So if our estimate is too large
	 * we need to decrease it until it is smaller.
	 */
	while(backmul.high > temp.high || backmul.high == temp.high && backmul.low > temp.low) {
	    sub128_128(&backmul, &divisor128);
	    uapprox--;
	}
	sub128_128(&temp, &backmul);
	result += uapprox;
    }
    *res = result;
    return 1;
}

int
nanosleep(const struct timespec *requested_time, struct timespec *remaining_time) {
    kern_return_t ret;
    uint64_t end, units;
    static struct mach_timebase_info info = {0, 0};
    static int unity;
    
    if ((requested_time == NULL) || (requested_time->tv_sec < 0) || (requested_time->tv_nsec > NSEC_PER_SEC)) {
        errno = EINVAL;
        return -1;
    }

    if (info.denom == 0) {
        ret = mach_timebase_info(&info);
        if (ret != KERN_SUCCESS) {
            fprintf(stderr, "mach_timebase_info() failed: %s\n", mach_error_string(ret));
            errno = EAGAIN;
            return -1;
        }
	/* If numer == denom == 1 (as in intel), no conversion needed */
	unity = (info.numer == info.denom);
    }

    if(unity)
	units = (uint64_t)requested_time->tv_sec * NSEC_PER_SEC;
    else if(!muldiv128((uint64_t)info.denom * NSEC_PER_SEC,
		       (uint64_t)requested_time->tv_sec,
		       (uint64_t)info.numer,
		       &units))
    {
	errno = EINVAL;
	return -1;
    }
    end = mach_absolute_time()
	+ units
	+ (uint64_t)info.denom * requested_time->tv_nsec / info.numer;
    ret = mach_wait_until(end);
    if (ret != KERN_SUCCESS) {
        if (ret == KERN_ABORTED) {
            errno = EINTR;
            if (remaining_time != NULL) {
                uint64_t now = mach_absolute_time();
                if (now >= end) {
		    remaining_time->tv_sec = 0;
		    remaining_time->tv_nsec = 0;
		} else {
		    if(unity)
			units = (end - now);
		    else
			muldiv128((uint64_t)info.numer,
				  (end - now),
				  (uint64_t)info.denom,
				  &units); // this can't overflow
		    remaining_time->tv_sec = units / NSEC_PER_SEC;
		    remaining_time->tv_nsec = units % NSEC_PER_SEC;
		}
            }
        } else {
            errno = EINVAL;
        }
        return -1;
    }
    return 0;
}


#endif /* __DARWIN_UNIX03 */
Commit	Line	Data
e9ce8d39	1	/*
224c7076	2	* Copyright (c) 1999, 2003, 2006, 2007 Apple Inc. All rights reserved.
e9ce8d39 A	3	*
	4	* @APPLE_LICENSE_HEADER_START@
	5	*
734aad71 A	6	* This file contains Original Code and/or Modifications of Original Code
	7	* as defined in and that are subject to the Apple Public Source License
	8	* Version 2.0 (the 'License'). You may not use this file except in
	9	* compliance with the License. Please obtain a copy of the License at
	10	* http://www.opensource.apple.com/apsl/ and read it before using this
	11	* file.
	12	*
	13	* The Original Code and all software distributed under the License are
	14	* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
e9ce8d39 A	15	* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
e9ce8d39 A	16	* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
734aad71 A	17	* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
	18	* Please see the License for the specific language governing rights and
	19	* limitations under the License.
e9ce8d39 A	20	*
	21	* @APPLE_LICENSE_HEADER_END@
	22	*/
	23
	24	#include <errno.h>
	25	#include <sys/time.h>
e9ce8d39	26	#include <mach/mach_error.h>
59e0d9fe	27	#include <mach/mach_time.h>
e9ce8d39 A	28	#include <stdio.h>
e9ce8d39 A	29
3d9156a7	30
224c7076	31	#if __DARWIN_UNIX03
3d9156a7	32	#include "pthread_internals.h"
224c7076	33	#include <mach/clock.h>
3d9156a7 A	34
	35	extern int __unix_conforming;
	36	extern mach_port_t clock_port;
	37	extern semaphore_t clock_sem;
224c7076 A	38	#ifdef VARIANT_CANCELABLE
	39	extern void _pthread_testcancel(pthread_t thread, int isconforming);
	40	extern int __semwait_signal(int cond_sem, int mutex_sem, int timeout, int relative, time_t tv_sec, __int32_t tv_nsec);
	41	#define SEMWAIT_SIGNAL __semwait_signal
	42	#else /* !VARIANT_CANCELABLE */
	43	extern int __semwait_signal_nocancel(int cond_sem, int mutex_sem, int timeout, int relative, time_t tv_sec, __int32_t tv_nsec);
	44	#define SEMWAIT_SIGNAL __semwait_signal_nocancel
	45	#endif /* VARIANT_CANCELABLE */
3d9156a7 A	46
	47	int
	48	nanosleep(const struct timespec requested_time, struct timespec remaining_time) {
	49	kern_return_t kret;
	50	int ret;
	51	mach_timespec_t remain;
	52	mach_timespec_t current;
	53
	54	if (__unix_conforming == 0)
	55	__unix_conforming = 1;
224c7076 A	56
	57	#ifdef VARIANT_CANCELABLE
	58	_pthread_testcancel(pthread_self(), 1);
	59	#endif /* VARIANT_CANCELABLE */
	60
3d9156a7 A	61	if ((requested_time == NULL) \|\| (requested_time->tv_sec < 0) \|\| (requested_time->tv_nsec >= NSEC_PER_SEC)) {
	62	errno = EINVAL;
	63	return -1;
	64	}
	65
224c7076	66
3d9156a7 A	67	if (remaining_time != NULL) {
	68	kret = clock_get_time(clock_port, &current);
	69	if (kret != KERN_SUCCESS) {
224c7076	70	fprintf(stderr, "clock_get_time() failed: %s\n", mach_error_string(kret));
3d9156a7 A	71	return -1;
	72	}
	73	}
224c7076	74	ret = SEMWAIT_SIGNAL(clock_sem, MACH_PORT_NULL, 1, 1, requested_time->tv_sec, requested_time->tv_nsec);
3d9156a7 A	75	if (ret < 0) {
	76	if (errno == ETIMEDOUT) {
	77	return 0;
	78	} else if (errno == EINTR) {
	79	if (remaining_time != NULL) {
	80	ret = clock_get_time(clock_port, &remain);
	81	if (ret != KERN_SUCCESS) {
	82	fprintf(stderr, "clock_get_time() failed: %s\n", mach_error_string(ret));
	83	return -1;
	84	}
	85	/* This depends on the layout of a mach_timespec_t and timespec_t being equivalent */
	86	ADD_MACH_TIMESPEC(&current, requested_time);
	87	SUB_MACH_TIMESPEC(&current, &remain);
	88	remaining_time->tv_sec = current.tv_sec;
	89	remaining_time->tv_nsec = current.tv_nsec;
	90	}
	91	} else {
	92	errno = EINVAL;
	93	}
	94	}
	95	return -1;
	96	}
	97
	98
224c7076 A	99	#else /* !__DARWIN_UNIX03 */
	100
	101	typedef struct {
	102	uint64_t high;
	103	uint64_t low;
	104	} uint128_t;
	105
	106	/* 128-bit addition: acc += add */
	107	static inline void
	108	add128_128(uint128_t acc, uint128_t add)
	109	{
	110	acc->high += add->high;
	111	acc->low += add->low;
	112	if(acc->low < add->low)
	113	acc->high++; // carry
	114	}
	115
	116	/* 128-bit subtraction: acc -= sub */
	117	static inline void
	118	sub128_128(uint128_t acc, uint128_t sub)
	119	{
	120	acc->high -= sub->high;
	121	if(acc->low < sub->low)
	122	acc->high--; // borrow
	123	acc->low -= sub->low;
	124	}
	125
	126	#define TWO64 (((double)(1ULL << 32)) * ((double)(1ULL << 32)))
	127
	128	static inline double
	129	uint128_double(uint128_t *u)
	130	{
	131	return TWO64 * u->high + u->low; // may loses precision
	132	}
	133
	134	/* 64x64 -> 128 bit multiplication */
	135	static inline void
	136	mul64x64(uint64_t x, uint64_t y, uint128_t *prod)
	137	{
	138	uint128_t add;
	139	/*
	140	* Split the two 64-bit multiplicands into 32-bit parts:
	141	* x => 2^32 * x1 + x2
	142	* y => 2^32 * y1 + y2
	143	*/
	144	uint32_t x1 = (uint32_t)(x >> 32);
	145	uint32_t x2 = (uint32_t)x;
	146	uint32_t y1 = (uint32_t)(y >> 32);
	147	uint32_t y2 = (uint32_t)y;
	148	/*
	149	* direct multiplication:
	150	* x * y => 2^64 * (x1 * y1) + 2^32 (x1 * y2 + x2 * y1) + (x2 * y2)
	151	* The first and last terms are direct assignmenet into the uint128_t
	152	* structure. Then we add the middle two terms separately, to avoid
	153	* 64-bit overflow. (We could use the Karatsuba algorithm to save
	154	* one multiply, but it is harder to deal with 64-bit overflows.)
	155	*/
	156	prod->high = (uint64_t)x1 * (uint64_t)y1;
	157	prod->low = (uint64_t)x2 * (uint64_t)y2;
	158	add.low = (uint64_t)x1 * (uint64_t)y2;
	159	add.high = (add.low >> 32);
	160	add.low <<= 32;
	161	add128_128(prod, &add);
	162	add.low = (uint64_t)x2 * (uint64_t)y1;
163	add.high = (add.low >> 32);
164	add.low <<= 32;
165	add128_128(prod, &add);
166	}
167
168	/* calculate (x * y / divisor), using 128-bit internal calculations */
169	static int
170	muldiv128(uint64_t x, uint64_t y, uint64_t divisor, uint64_t *res)
171	{
172	uint128_t temp;
173	uint128_t divisor128 = {0, divisor};
174	uint64_t result = 0;
175	double recip;
176
177	/* calculate (x * y) */
178	mul64x64(x, y, &temp);
179	/*
180	* Now divide by the divisor. We use floating point to calculate an
181	* approximate answer and update the results. Then we iterate and
182	* calculate a correction from the difference.
183	*/
184	recip = 1.0 / ((double)divisor);
185	while(temp.high \|\| temp.low >= divisor) {
186	uint128_t backmul;
187	uint64_t uapprox;
188	double approx = uint128_double(&temp) * recip;
189
190	if(approx > __LONG_LONG_MAX__)
191	return 0; // answer overflows 64-bits
192	uapprox = (uint64_t)approx;
193	mul64x64(uapprox, divisor, &backmul);
194	/*
195	* Because we are using unsigned integers, we need to approach the
196	* answer from the lesser side. So if our estimate is too large
197	* we need to decrease it until it is smaller.
198	*/
199	while(backmul.high > temp.high \|\| backmul.high == temp.high && backmul.low > temp.low) {
200	sub128_128(&backmul, &divisor128);
201	uapprox--;
202	}
203	sub128_128(&temp, &backmul);
204	result += uapprox;
205	}
206	*res = result;
207	return 1;
208	}
3d9156a7	209
e9ce8d39 A	210	int
	211	nanosleep(const struct timespec requested_time, struct timespec remaining_time) {
	212	kern_return_t ret;
224c7076 A	213	uint64_t end, units;
	214	static struct mach_timebase_info info = {0, 0};
	215	static int unity;
e9ce8d39 A	216
	217	if ((requested_time == NULL) \|\| (requested_time->tv_sec < 0) \|\| (requested_time->tv_nsec > NSEC_PER_SEC)) {
	218	errno = EINVAL;
	219	return -1;
	220	}
	221
224c7076	222	if (info.denom == 0) {
59e0d9fe A	223	ret = mach_timebase_info(&info);
	224	if (ret != KERN_SUCCESS) {
	225	fprintf(stderr, "mach_timebase_info() failed: %s\n", mach_error_string(ret));
	226	errno = EAGAIN;
	227	return -1;
	228	}
224c7076 A	229	/* If numer == denom == 1 (as in intel), no conversion needed */
224c7076 A	230	unity = (info.numer == info.denom);
e9ce8d39	231	}
59e0d9fe	232
224c7076 A	233	if(unity)
	234	units = (uint64_t)requested_time->tv_sec * NSEC_PER_SEC;
	235	else if(!muldiv128((uint64_t)info.denom * NSEC_PER_SEC,
	236	(uint64_t)requested_time->tv_sec,
	237	(uint64_t)info.numer,
	238	&units))
	239	{
	240	errno = EINVAL;
	241	return -1;
	242	}
	243	end = mach_absolute_time()
	244	+ units
	245	+ (uint64_t)info.denom * requested_time->tv_nsec / info.numer;
59e0d9fe	246	ret = mach_wait_until(end);
e9ce8d39 A	247	if (ret != KERN_SUCCESS) {
	248	if (ret == KERN_ABORTED) {
	249	errno = EINTR;
	250	if (remaining_time != NULL) {
59e0d9fe	251	uint64_t now = mach_absolute_time();
224c7076 A	252	if (now >= end) {
	253	remaining_time->tv_sec = 0;
	254	remaining_time->tv_nsec = 0;
	255	} else {
	256	if(unity)
	257	units = (end - now);
	258	else
	259	muldiv128((uint64_t)info.numer,
	260	(end - now),
	261	(uint64_t)info.denom,
	262	&units); // this can't overflow
	263	remaining_time->tv_sec = units / NSEC_PER_SEC;
	264	remaining_time->tv_nsec = units % NSEC_PER_SEC;
	265	}
e9ce8d39 A	266	}
	267	} else {
	268	errno = EINVAL;
	269	}
	270	return -1;
	271	}
	272	return 0;
	273	}
3d9156a7 A	274
3d9156a7 A	275
224c7076	276	#endif /* __DARWIN_UNIX03 */