[thirdparty/openssl.git] / crypto / bn / bn_div.c

/*
 * Copyright 1995-2017 The OpenSSL Project Authors. All Rights Reserved.
 *
 * Licensed under the OpenSSL license (the "License").  You may not use
 * this file except in compliance with the License.  You can obtain a copy
 * in the file LICENSE in the source distribution or at
 * https://www.openssl.org/source/license.html
 */

#include <openssl/bn.h>
#include "internal/cryptlib.h"
#include "bn_lcl.h"

/* The old slow way */
#if 0
int BN_div(BIGNUM *dv, BIGNUM *rem, const BIGNUM *m, const BIGNUM *d,
           BN_CTX *ctx)
{
    int i, nm, nd;
    int ret = 0;
    BIGNUM *D;

    bn_check_top(m);
    bn_check_top(d);
    if (BN_is_zero(d)) {
        BNerr(BN_F_BN_DIV, BN_R_DIV_BY_ZERO);
        return 0;
    }

    if (BN_ucmp(m, d) < 0) {
        if (rem != NULL) {
            if (BN_copy(rem, m) == NULL)
                return 0;
        }
        if (dv != NULL)
            BN_zero(dv);
        return 1;
    }

    BN_CTX_start(ctx);
    D = BN_CTX_get(ctx);
    if (dv == NULL)
        dv = BN_CTX_get(ctx);
    if (rem == NULL)
        rem = BN_CTX_get(ctx);
    if (D == NULL || dv == NULL || rem == NULL)
        goto end;

    nd = BN_num_bits(d);
    nm = BN_num_bits(m);
    if (BN_copy(D, d) == NULL)
        goto end;
    if (BN_copy(rem, m) == NULL)
        goto end;

    /*
     * The next 2 are needed so we can do a dv->d[0]|=1 later since
     * BN_lshift1 will only work once there is a value :-)
     */
    BN_zero(dv);
    if (bn_wexpand(dv, 1) == NULL)
        goto end;
    dv->top = 1;

    if (!BN_lshift(D, D, nm - nd))
        goto end;
    for (i = nm - nd; i >= 0; i--) {
        if (!BN_lshift1(dv, dv))
            goto end;
        if (BN_ucmp(rem, D) >= 0) {
            dv->d[0] |= 1;
            if (!BN_usub(rem, rem, D))
                goto end;
        }
/* CAN IMPROVE (and have now :=) */
        if (!BN_rshift1(D, D))
            goto end;
    }
    rem->neg = BN_is_zero(rem) ? 0 : m->neg;
    dv->neg = m->neg ^ d->neg;
    ret = 1;
 end:
    BN_CTX_end(ctx);
    return ret;
}

#else

# if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \
    && !defined(PEDANTIC) && !defined(BN_DIV3W)
#  if defined(__GNUC__) && __GNUC__>=2
#   if defined(__i386) || defined (__i386__)
   /*-
    * There were two reasons for implementing this template:
    * - GNU C generates a call to a function (__udivdi3 to be exact)
    *   in reply to ((((BN_ULLONG)n0)<<BN_BITS2)|n1)/d0 (I fail to
    *   understand why...);
    * - divl doesn't only calculate quotient, but also leaves
    *   remainder in %edx which we can definitely use here:-)
    */
#    undef bn_div_words
#    define bn_div_words(n0,n1,d0)                \
        ({  asm volatile (                      \
                "divl   %4"                     \
                : "=a"(q), "=d"(rem)            \
                : "a"(n1), "d"(n0), "r"(d0)     \
                : "cc");                        \
            q;                                  \
        })
#    define REMAINDER_IS_ALREADY_CALCULATED
#   elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
   /*
    * Same story here, but it's 128-bit by 64-bit division. Wow!
    */
#    undef bn_div_words
#    define bn_div_words(n0,n1,d0)                \
        ({  asm volatile (                      \
                "divq   %4"                     \
                : "=a"(q), "=d"(rem)            \
                : "a"(n1), "d"(n0), "r"(d0)     \
                : "cc");                        \
            q;                                  \
        })
#    define REMAINDER_IS_ALREADY_CALCULATED
#   endif                       /* __<cpu> */
#  endif                        /* __GNUC__ */
# endif                         /* OPENSSL_NO_ASM */

/*-
 * BN_div computes  dv := num / divisor, rounding towards
 * zero, and sets up rm  such that  dv*divisor + rm = num  holds.
 * Thus:
 *     dv->neg == num->neg ^ divisor->neg  (unless the result is zero)
 *     rm->neg == num->neg                 (unless the remainder is zero)
 * If 'dv' or 'rm' is NULL, the respective value is not returned.
 */
int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
           BN_CTX *ctx)
{
    int norm_shift, i, loop;
    BIGNUM *tmp, wnum, *snum, *sdiv, *res;
    BN_ULONG *resp, *wnump;
    BN_ULONG d0, d1;
    int num_n, div_n;
    int no_branch = 0;

    /*
     * Invalid zero-padding would have particularly bad consequences so don't
     * just rely on bn_check_top() here (bn_check_top() works only for
     * BN_DEBUG builds)
     */
    if ((num->top > 0 && num->d[num->top - 1] == 0) ||
        (divisor->top > 0 && divisor->d[divisor->top - 1] == 0)) {
        BNerr(BN_F_BN_DIV, BN_R_NOT_INITIALIZED);
        return 0;
    }

    bn_check_top(num);
    bn_check_top(divisor);

    if ((BN_get_flags(num, BN_FLG_CONSTTIME) != 0)
        || (BN_get_flags(divisor, BN_FLG_CONSTTIME) != 0)) {
        no_branch = 1;
    }

    bn_check_top(dv);
    bn_check_top(rm);
    /*- bn_check_top(num); *//*
     * 'num' has been checked already
     */
    /*- bn_check_top(divisor); *//*
     * 'divisor' has been checked already
     */

    if (BN_is_zero(divisor)) {
        BNerr(BN_F_BN_DIV, BN_R_DIV_BY_ZERO);
        return 0;
    }

    if (!no_branch && BN_ucmp(num, divisor) < 0) {
        if (rm != NULL) {
            if (BN_copy(rm, num) == NULL)
                return 0;
        }
        if (dv != NULL)
            BN_zero(dv);
        return 1;
    }

    BN_CTX_start(ctx);
    res = (dv == NULL) ? BN_CTX_get(ctx) : dv;
    tmp = BN_CTX_get(ctx);
    snum = BN_CTX_get(ctx);
    sdiv = BN_CTX_get(ctx);
    if (sdiv == NULL)
        goto err;

    /* First we normalise the numbers */
    norm_shift = BN_BITS2 - ((BN_num_bits(divisor)) % BN_BITS2);
    if (!(BN_lshift(sdiv, divisor, norm_shift)))
        goto err;
    sdiv->neg = 0;
    norm_shift += BN_BITS2;
    if (!(BN_lshift(snum, num, norm_shift)))
        goto err;
    snum->neg = 0;

    if (no_branch) {
        /*
         * Since we don't know whether snum is larger than sdiv, we pad snum
         * with enough zeroes without changing its value.
         */
        if (snum->top <= sdiv->top + 1) {
            if (bn_wexpand(snum, sdiv->top + 2) == NULL)
                goto err;
            for (i = snum->top; i < sdiv->top + 2; i++)
                snum->d[i] = 0;
            snum->top = sdiv->top + 2;
        } else {
            if (bn_wexpand(snum, snum->top + 1) == NULL)
                goto err;
            snum->d[snum->top] = 0;
            snum->top++;
        }
    }

    div_n = sdiv->top;
    num_n = snum->top;
    loop = num_n - div_n;
    /*
     * Lets setup a 'window' into snum This is the part that corresponds to
     * the current 'area' being divided
     */
    wnum.neg = 0;
    wnum.d = &(snum->d[loop]);
    wnum.top = div_n;
    /*
     * only needed when BN_ucmp messes up the values between top and max
     */
    wnum.dmax = snum->dmax - loop; /* so we don't step out of bounds */

    /* Get the top 2 words of sdiv */
    /* div_n=sdiv->top; */
    d0 = sdiv->d[div_n - 1];
    d1 = (div_n == 1) ? 0 : sdiv->d[div_n - 2];

    /* pointer to the 'top' of snum */
    wnump = &(snum->d[num_n - 1]);

    /* Setup to 'res' */
    if (!bn_wexpand(res, (loop + 1)))
        goto err;
    res->neg = (num->neg ^ divisor->neg);
    res->top = loop - no_branch;
    resp = &(res->d[loop - 1]);

    /* space for temp */
    if (!bn_wexpand(tmp, (div_n + 1)))
        goto err;

    if (!no_branch) {
        if (BN_ucmp(&wnum, sdiv) >= 0) {
            /*
             * If BN_DEBUG_RAND is defined BN_ucmp changes (via bn_pollute)
             * the const bignum arguments => clean the values between top and
             * max again
             */
            bn_clear_top2max(&wnum);
            bn_sub_words(wnum.d, wnum.d, sdiv->d, div_n);
            *resp = 1;
        } else
            res->top--;
    }

    /* Increase the resp pointer so that we never create an invalid pointer. */
    resp++;

    /*
     * if res->top == 0 then clear the neg value otherwise decrease the resp
     * pointer
     */
    if (res->top == 0)
        res->neg = 0;
    else
        resp--;

    for (i = 0; i < loop - 1; i++, wnump--) {
        BN_ULONG q, l0;
        /*
         * the first part of the loop uses the top two words of snum and sdiv
         * to calculate a BN_ULONG q such that | wnum - sdiv * q | < sdiv
         */
# if defined(BN_DIV3W) && !defined(OPENSSL_NO_ASM)
        BN_ULONG bn_div_3_words(BN_ULONG *, BN_ULONG, BN_ULONG);
        q = bn_div_3_words(wnump, d1, d0);
# else
        BN_ULONG n0, n1, rem = 0;

        n0 = wnump[0];
        n1 = wnump[-1];
        if (n0 == d0)
            q = BN_MASK2;
        else {                  /* n0 < d0 */

#  ifdef BN_LLONG
            BN_ULLONG t2;

#   if defined(BN_LLONG) && defined(BN_DIV2W) && !defined(bn_div_words)
            q = (BN_ULONG)(((((BN_ULLONG) n0) << BN_BITS2) | n1) / d0);
#   else
            q = bn_div_words(n0, n1, d0);
#   endif

#   ifndef REMAINDER_IS_ALREADY_CALCULATED
            /*
             * rem doesn't have to be BN_ULLONG. The least we
             * know it's less that d0, isn't it?
             */
            rem = (n1 - q * d0) & BN_MASK2;
#   endif
            t2 = (BN_ULLONG) d1 *q;

            for (;;) {
                if (t2 <= ((((BN_ULLONG) rem) << BN_BITS2) | wnump[-2]))
                    break;
                q--;
                rem += d0;
                if (rem < d0)
                    break;      /* don't let rem overflow */
                t2 -= d1;
            }
#  else                         /* !BN_LLONG */
            BN_ULONG t2l, t2h;

            q = bn_div_words(n0, n1, d0);
#   ifndef REMAINDER_IS_ALREADY_CALCULATED
            rem = (n1 - q * d0) & BN_MASK2;
#   endif

#   if defined(BN_UMULT_LOHI)
            BN_UMULT_LOHI(t2l, t2h, d1, q);
#   elif defined(BN_UMULT_HIGH)
            t2l = d1 * q;
            t2h = BN_UMULT_HIGH(d1, q);
#   else
            {
                BN_ULONG ql, qh;
                t2l = LBITS(d1);
                t2h = HBITS(d1);
                ql = LBITS(q);
                qh = HBITS(q);
                mul64(t2l, t2h, ql, qh); /* t2=(BN_ULLONG)d1*q; */
            }
#   endif

            for (;;) {
                if ((t2h < rem) || ((t2h == rem) && (t2l <= wnump[-2])))
                    break;
                q--;
                rem += d0;
                if (rem < d0)
                    break;      /* don't let rem overflow */
                if (t2l < d1)
                    t2h--;
                t2l -= d1;
            }
#  endif                        /* !BN_LLONG */
        }
# endif                         /* !BN_DIV3W */

        l0 = bn_mul_words(tmp->d, sdiv->d, div_n, q);
        tmp->d[div_n] = l0;
        wnum.d--;
        /*
         * ingore top values of the bignums just sub the two BN_ULONG arrays
         * with bn_sub_words
         */
        if (bn_sub_words(wnum.d, wnum.d, tmp->d, div_n + 1)) {
            /*
             * Note: As we have considered only the leading two BN_ULONGs in
             * the calculation of q, sdiv * q might be greater than wnum (but
             * then (q-1) * sdiv is less or equal than wnum)
             */
            q--;
            if (bn_add_words(wnum.d, wnum.d, sdiv->d, div_n))
                /*
                 * we can't have an overflow here (assuming that q != 0, but
                 * if q == 0 then tmp is zero anyway)
                 */
                (*wnump)++;
        }
        /* store part of the result */
        resp--;
        *resp = q;
    }
    bn_correct_top(snum);
    if (rm != NULL) {
        /*
         * Keep a copy of the neg flag in num because if rm==num BN_rshift()
         * will overwrite it.
         */
        int neg = num->neg;
        BN_rshift(rm, snum, norm_shift);
        if (!BN_is_zero(rm))
            rm->neg = neg;
        bn_check_top(rm);
    }
    if (no_branch)
        bn_correct_top(res);
    BN_CTX_end(ctx);
    return 1;
 err:
    bn_check_top(rm);
    BN_CTX_end(ctx);
    return 0;
}
#endif
Commit	Line	Data
4f22f405	1	/*
edea42c6	2	* Copyright 1995-2017 The OpenSSL Project Authors. All Rights Reserved.
d02b48c6	3	*
4f22f405 RS	4	* Licensed under the OpenSSL license (the "License"). You may not use
	5	* this file except in compliance with the License. You can obtain a copy
	6	* in the file LICENSE in the source distribution or at
	7	* https://www.openssl.org/source/license.html
d02b48c6 RE	8	*/
d02b48c6 RE	9
84c15db5	10	#include <openssl/bn.h>
b39fc560	11	#include "internal/cryptlib.h"
d02b48c6 RE	12	#include "bn_lcl.h"
	13
	14	/* The old slow way */
4a6222d7	15	#if 0
0bde1089	16	int BN_div(BIGNUM dv, BIGNUM rem, const BIGNUM m, const BIGNUM d,
0f113f3e MC	17	BN_CTX *ctx)
	18	{
	19	int i, nm, nd;
	20	int ret = 0;
	21	BIGNUM *D;
	22
	23	bn_check_top(m);
	24	bn_check_top(d);
	25	if (BN_is_zero(d)) {
	26	BNerr(BN_F_BN_DIV, BN_R_DIV_BY_ZERO);
26a7d938	27	return 0;
0f113f3e MC	28	}
	29
	30	if (BN_ucmp(m, d) < 0) {
	31	if (rem != NULL) {
	32	if (BN_copy(rem, m) == NULL)
26a7d938	33	return 0;
0f113f3e MC	34	}
	35	if (dv != NULL)
	36	BN_zero(dv);
208fb891	37	return 1;
0f113f3e MC	38	}
	39
	40	BN_CTX_start(ctx);
	41	D = BN_CTX_get(ctx);
	42	if (dv == NULL)
	43	dv = BN_CTX_get(ctx);
	44	if (rem == NULL)
	45	rem = BN_CTX_get(ctx);
	46	if (D == NULL \|\| dv == NULL \|\| rem == NULL)
	47	goto end;
	48
	49	nd = BN_num_bits(d);
	50	nm = BN_num_bits(m);
	51	if (BN_copy(D, d) == NULL)
	52	goto end;
	53	if (BN_copy(rem, m) == NULL)
	54	goto end;
	55
	56	/*
	57	* The next 2 are needed so we can do a dv->d[0]\|=1 later since
	58	* BN_lshift1 will only work once there is a value :-)
	59	*/
	60	BN_zero(dv);
	61	if (bn_wexpand(dv, 1) == NULL)
	62	goto end;
	63	dv->top = 1;
	64
	65	if (!BN_lshift(D, D, nm - nd))
	66	goto end;
	67	for (i = nm - nd; i >= 0; i--) {
	68	if (!BN_lshift1(dv, dv))
	69	goto end;
	70	if (BN_ucmp(rem, D) >= 0) {
	71	dv->d[0] \|= 1;
	72	if (!BN_usub(rem, rem, D))
	73	goto end;
	74	}
d02b48c6	75	/* CAN IMPROVE (and have now :=) */
0f113f3e MC	76	if (!BN_rshift1(D, D))
	77	goto end;
	78	}
	79	rem->neg = BN_is_zero(rem) ? 0 : m->neg;
	80	dv->neg = m->neg ^ d->neg;
	81	ret = 1;
9b141126	82	end:
0f113f3e	83	BN_CTX_end(ctx);
26a7d938	84	return ret;
0f113f3e	85	}
d02b48c6 RE	86
	87	#else
	88
0f113f3e	89	# if !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) \
cf1b7d96	90	&& !defined(PEDANTIC) && !defined(BN_DIV3W)
0f113f3e MC	91	# if defined(__GNUC__) && __GNUC__>=2
0f113f3e MC	92	# if defined(__i386) \|\| defined (__i386__)
c80fd6b2	93	/*-
4a6222d7 UM	94	* There were two reasons for implementing this template:
	95	* - GNU C generates a call to a function (__udivdi3 to be exact)
	96	* in reply to ((((BN_ULLONG)n0)<<BN_BITS2)\|n1)/d0 (I fail to
	97	* understand why...);
	98	* - divl doesn't only calculate quotient, but also leaves
	99	* remainder in %edx which we can definitely use here:-)
4a6222d7	100	*/
0f113f3e MC	101	# undef bn_div_words
	102	# define bn_div_words(n0,n1,d0) \
	103	({ asm volatile ( \
	104	"divl %4" \
	105	: "=a"(q), "=d"(rem) \
68b4a6e9	106	: "a"(n1), "d"(n0), "r"(d0) \
0f113f3e MC	107	: "cc"); \
	108	q; \
	109	})
	110	# define REMAINDER_IS_ALREADY_CALCULATED
	111	# elif defined(__x86_64) && defined(SIXTY_FOUR_BIT_LONG)
2f98abbc AP	112	/*
2f98abbc AP	113	* Same story here, but it's 128-bit by 64-bit division. Wow!
2f98abbc	114	*/
0f113f3e MC	115	# undef bn_div_words
	116	# define bn_div_words(n0,n1,d0) \
	117	({ asm volatile ( \
	118	"divq %4" \
	119	: "=a"(q), "=d"(rem) \
68b4a6e9	120	: "a"(n1), "d"(n0), "r"(d0) \
0f113f3e MC	121	: "cc"); \
	122	q; \
	123	})
	124	# define REMAINDER_IS_ALREADY_CALCULATED
	125	# endif /* __<cpu> */
	126	# endif /* __GNUC__ */
	127	# endif /* OPENSSL_NO_ASM */
78a0c1f1	128
1d97c843	129	/*-
02e112a8	130	* BN_div computes dv := num / divisor, rounding towards
55525742	131	* zero, and sets up rm such that dv*divisor + rm = num holds.
78a0c1f1 BM	132	* Thus:
	133	* dv->neg == num->neg ^ divisor->neg (unless the result is zero)
	134	* rm->neg == num->neg (unless the remainder is zero)
	135	* If 'dv' or 'rm' is NULL, the respective value is not returned.
	136	*/
84c15db5	137	int BN_div(BIGNUM dv, BIGNUM rm, const BIGNUM num, const BIGNUM divisor,
0f113f3e MC	138	BN_CTX *ctx)
	139	{
	140	int norm_shift, i, loop;
	141	BIGNUM tmp, wnum, snum, sdiv, res;
	142	BN_ULONG resp, wnump;
	143	BN_ULONG d0, d1;
	144	int num_n, div_n;
	145	int no_branch = 0;
	146
	147	/*
	148	* Invalid zero-padding would have particularly bad consequences so don't
	149	* just rely on bn_check_top() here (bn_check_top() works only for
	150	* BN_DEBUG builds)
	151	*/
	152	if ((num->top > 0 && num->d[num->top - 1] == 0) \|\|
	153	(divisor->top > 0 && divisor->d[divisor->top - 1] == 0)) {
	154	BNerr(BN_F_BN_DIV, BN_R_NOT_INITIALIZED);
	155	return 0;
	156	}
	157
	158	bn_check_top(num);
	159	bn_check_top(divisor);
	160
	161	if ((BN_get_flags(num, BN_FLG_CONSTTIME) != 0)
	162	\|\| (BN_get_flags(divisor, BN_FLG_CONSTTIME) != 0)) {
	163	no_branch = 1;
	164	}
	165
	166	bn_check_top(dv);
	167	bn_check_top(rm);
	168	/- bn_check_top(num); //*
	169	* 'num' has been checked already
	170	*/
	171	/- bn_check_top(divisor); //*
	172	* 'divisor' has been checked already
	173	*/
	174
	175	if (BN_is_zero(divisor)) {
	176	BNerr(BN_F_BN_DIV, BN_R_DIV_BY_ZERO);
26a7d938	177	return 0;
0f113f3e MC	178	}
	179
	180	if (!no_branch && BN_ucmp(num, divisor) < 0) {
	181	if (rm != NULL) {
	182	if (BN_copy(rm, num) == NULL)
26a7d938	183	return 0;
0f113f3e MC	184	}
	185	if (dv != NULL)
	186	BN_zero(dv);
208fb891	187	return 1;
0f113f3e MC	188	}
	189
	190	BN_CTX_start(ctx);
edea42c6	191	res = (dv == NULL) ? BN_CTX_get(ctx) : dv;
0f113f3e MC	192	tmp = BN_CTX_get(ctx);
	193	snum = BN_CTX_get(ctx);
	194	sdiv = BN_CTX_get(ctx);
edea42c6	195	if (sdiv == NULL)
0f113f3e MC	196	goto err;
	197
	198	/* First we normalise the numbers */
	199	norm_shift = BN_BITS2 - ((BN_num_bits(divisor)) % BN_BITS2);
	200	if (!(BN_lshift(sdiv, divisor, norm_shift)))
	201	goto err;
	202	sdiv->neg = 0;
	203	norm_shift += BN_BITS2;
	204	if (!(BN_lshift(snum, num, norm_shift)))
	205	goto err;
	206	snum->neg = 0;
	207
	208	if (no_branch) {
	209	/*
	210	* Since we don't know whether snum is larger than sdiv, we pad snum
	211	* with enough zeroes without changing its value.
	212	*/
	213	if (snum->top <= sdiv->top + 1) {
	214	if (bn_wexpand(snum, sdiv->top + 2) == NULL)
	215	goto err;
	216	for (i = snum->top; i < sdiv->top + 2; i++)
	217	snum->d[i] = 0;
	218	snum->top = sdiv->top + 2;
	219	} else {
	220	if (bn_wexpand(snum, snum->top + 1) == NULL)
	221	goto err;
	222	snum->d[snum->top] = 0;
	223	snum->top++;
	224	}
	225	}
	226
	227	div_n = sdiv->top;
	228	num_n = snum->top;
	229	loop = num_n - div_n;
	230	/*
	231	* Lets setup a 'window' into snum This is the part that corresponds to
	232	* the current 'area' being divided
	233	*/
	234	wnum.neg = 0;
	235	wnum.d = &(snum->d[loop]);
	236	wnum.top = div_n;
	237	/*
	238	* only needed when BN_ucmp messes up the values between top and max
	239	*/
	240	wnum.dmax = snum->dmax - loop; /* so we don't step out of bounds */
	241
	242	/* Get the top 2 words of sdiv */
	243	/* div_n=sdiv->top; */
	244	d0 = sdiv->d[div_n - 1];
	245	d1 = (div_n == 1) ? 0 : sdiv->d[div_n - 2];
	246
	247	/* pointer to the 'top' of snum */
	248	wnump = &(snum->d[num_n - 1]);
	249
	250	/* Setup to 'res' */
0f113f3e MC	251	if (!bn_wexpand(res, (loop + 1)))
0f113f3e MC	252	goto err;
38d1b3cc	253	res->neg = (num->neg ^ divisor->neg);
0f113f3e MC	254	res->top = loop - no_branch;
	255	resp = &(res->d[loop - 1]);
	256
	257	/* space for temp */
	258	if (!bn_wexpand(tmp, (div_n + 1)))
	259	goto err;
	260
	261	if (!no_branch) {
	262	if (BN_ucmp(&wnum, sdiv) >= 0) {
	263	/*
	264	* If BN_DEBUG_RAND is defined BN_ucmp changes (via bn_pollute)
	265	* the const bignum arguments => clean the values between top and
	266	* max again
	267	*/
	268	bn_clear_top2max(&wnum);
	269	bn_sub_words(wnum.d, wnum.d, sdiv->d, div_n);
	270	*resp = 1;
	271	} else
	272	res->top--;
	273	}
	274
acc60092 KR	275	/* Increase the resp pointer so that we never create an invalid pointer. */
	276	resp++;
	277
0f113f3e MC	278	/*
	279	* if res->top == 0 then clear the neg value otherwise decrease the resp
	280	* pointer
	281	*/
	282	if (res->top == 0)
	283	res->neg = 0;
	284	else
	285	resp--;
	286
acc60092	287	for (i = 0; i < loop - 1; i++, wnump--) {
0f113f3e MC	288	BN_ULONG q, l0;
	289	/*
	290	* the first part of the loop uses the top two words of snum and sdiv
	291	* to calculate a BN_ULONG q such that \| wnum - sdiv * q \| < sdiv
	292	*/
	293	# if defined(BN_DIV3W) && !defined(OPENSSL_NO_ASM)
	294	BN_ULONG bn_div_3_words(BN_ULONG *, BN_ULONG, BN_ULONG);
	295	q = bn_div_3_words(wnump, d1, d0);
	296	# else
	297	BN_ULONG n0, n1, rem = 0;
	298
	299	n0 = wnump[0];
	300	n1 = wnump[-1];
	301	if (n0 == d0)
	302	q = BN_MASK2;
	303	else { /* n0 < d0 */
	304
	305	# ifdef BN_LLONG
	306	BN_ULLONG t2;
	307
	308	# if defined(BN_LLONG) && defined(BN_DIV2W) && !defined(bn_div_words)
	309	q = (BN_ULONG)(((((BN_ULLONG) n0) << BN_BITS2) \| n1) / d0);
	310	# else
	311	q = bn_div_words(n0, n1, d0);
0f113f3e MC	312	# endif
	313
	314	# ifndef REMAINDER_IS_ALREADY_CALCULATED
	315	/*
	316	* rem doesn't have to be BN_ULLONG. The least we
	317	* know it's less that d0, isn't it?
	318	*/
	319	rem = (n1 - q * d0) & BN_MASK2;
	320	# endif
	321	t2 = (BN_ULLONG) d1 *q;
	322
	323	for (;;) {
	324	if (t2 <= ((((BN_ULLONG) rem) << BN_BITS2) \| wnump[-2]))
	325	break;
	326	q--;
	327	rem += d0;
	328	if (rem < d0)
	329	break; /* don't let rem overflow */
	330	t2 -= d1;
	331	}
	332	# else /* !BN_LLONG */
	333	BN_ULONG t2l, t2h;
	334
	335	q = bn_div_words(n0, n1, d0);
0f113f3e MC	336	# ifndef REMAINDER_IS_ALREADY_CALCULATED
	337	rem = (n1 - q * d0) & BN_MASK2;
	338	# endif
	339
	340	# if defined(BN_UMULT_LOHI)
	341	BN_UMULT_LOHI(t2l, t2h, d1, q);
	342	# elif defined(BN_UMULT_HIGH)
	343	t2l = d1 * q;
	344	t2h = BN_UMULT_HIGH(d1, q);
	345	# else
	346	{
	347	BN_ULONG ql, qh;
	348	t2l = LBITS(d1);
	349	t2h = HBITS(d1);
	350	ql = LBITS(q);
	351	qh = HBITS(q);
	352	mul64(t2l, t2h, ql, qh); /* t2=(BN_ULLONG)d1q; /
	353	}
	354	# endif
	355
	356	for (;;) {
	357	if ((t2h < rem) \|\| ((t2h == rem) && (t2l <= wnump[-2])))
	358	break;
	359	q--;
	360	rem += d0;
	361	if (rem < d0)
	362	break; /* don't let rem overflow */
	363	if (t2l < d1)
	364	t2h--;
	365	t2l -= d1;
	366	}
	367	# endif /* !BN_LLONG */
	368	}
	369	# endif /* !BN_DIV3W */
	370
	371	l0 = bn_mul_words(tmp->d, sdiv->d, div_n, q);
	372	tmp->d[div_n] = l0;
	373	wnum.d--;
	374	/*
	375	* ingore top values of the bignums just sub the two BN_ULONG arrays
	376	* with bn_sub_words
	377	*/
	378	if (bn_sub_words(wnum.d, wnum.d, tmp->d, div_n + 1)) {
	379	/*
	380	* Note: As we have considered only the leading two BN_ULONGs in
	381	* the calculation of q, sdiv * q might be greater than wnum (but
	382	* then (q-1) * sdiv is less or equal than wnum)
	383	*/
	384	q--;
	385	if (bn_add_words(wnum.d, wnum.d, sdiv->d, div_n))
	386	/*
	387	* we can't have an overflow here (assuming that q != 0, but
	388	* if q == 0 then tmp is zero anyway)
	389	*/
	390	(*wnump)++;
	391	}
	392	/* store part of the result */
acc60092	393	resp--;
0f113f3e MC	394	*resp = q;
	395	}
	396	bn_correct_top(snum);
	397	if (rm != NULL) {
	398	/*
	399	* Keep a copy of the neg flag in num because if rm==num BN_rshift()
	400	* will overwrite it.
	401	*/
	402	int neg = num->neg;
	403	BN_rshift(rm, snum, norm_shift);
	404	if (!BN_is_zero(rm))
	405	rm->neg = neg;
	406	bn_check_top(rm);
	407	}
	408	if (no_branch)
	409	bn_correct_top(res);
	410	BN_CTX_end(ctx);
208fb891	411	return 1;
0f113f3e MC	412	err:
	413	bn_check_top(rm);
	414	BN_CTX_end(ctx);
26a7d938	415	return 0;
0f113f3e	416	}
d02b48c6	417	#endif