Exemple #1
0
int BN_div(BIGNUM *dv, BIGNUM *rm, const BIGNUM *num, const BIGNUM *divisor,
	   BN_CTX *ctx)
	{
	int norm_shift,i,loop;
	BIGNUM *tmp,*snum,*sdiv,*res;
#ifdef BN_ALLOC				/* pcg */
	BIGNUM wnum;
#else
	BN_ULONG *wnum_d;
#endif /* !BN_ALLOC */
	BN_ULONG *resp,*wnump;
	BN_ULONG d0,d1;
	int num_n,div_n;

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

	bn_check_top(num);

	if ((BN_get_flags(num, BN_FLG_CONSTTIME) != 0) || (BN_get_flags(divisor, BN_FLG_CONSTTIME) != 0))
		{
		return BN_div_no_branch(dv, rm, num, divisor, ctx);
		}

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

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

	if (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);
	tmp=BN_CTX_get(ctx);
	snum=BN_CTX_get_ext( ctx, BIGNUM_EXT_MUL1 );	/* pcg */
	sdiv=BN_CTX_get(ctx);
	if (dv == NULL)
		res=BN_CTX_get(ctx);
	else	res=dv;
	if (sdiv == NULL || res == NULL || tmp == NULL || snum == 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;
	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 */
#ifdef BN_ALLOC				/* pcg */
	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 */
#else
	wnum_d = &(snum->d[loop]);
#endif /* BN_ALLOC */

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

	/* space for temp */
	if( div_n+1 > BIGNUM_ALLOC_WORDS ) { assert( 0 ); goto err; }

	if (BN_ucmp_word(wnum_d,div_n,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--;
	/* 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--, resp--)
		{
		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);
#ifdef BN_DEBUG_LEVITTE
			fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
X) -> 0x%08X\n",
				n0, n1, d0, q);
#endif
#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;
#if !defined( BN_UMULT_LOHI ) && !defined( BN_UMULT_HIGH )	/* pcg */
			BN_ULONG ql,qh;
#endif 

			q=bn_div_words(n0,n1,d0);
#ifdef BN_DEBUG_LEVITTE
			fprintf(stderr,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
X) -> 0x%08X\n",
				n0, n1, d0, q);
#endif
#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 = 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);
		}
	BN_CTX_end_ext( ctx, BIGNUM_EXT_MUL1 );		/* pcg */
	return(1);
err:
	bn_check_top(rm);
	BN_CTX_end_ext( ctx, BIGNUM_EXT_MUL1 );		/* pcg */
	return(0);
	}
/* BN_div_no_branch is a special version of BN_div. It does not contain
 * branches that may leak sensitive information.
 */
static int BN_div_no_branch(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;

	bn_check_top(dv);
	bn_check_top(rm);
	/* bn_check_top(num); */ /* 'num' has been checked in BN_div() */
	bn_check_top(divisor);

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

	BN_CTX_start(ctx);
	tmp=BN_CTX_get(ctx);
	snum=BN_CTX_get(ctx);
	sdiv=BN_CTX_get(ctx);
	if (dv == NULL)
		res=BN_CTX_get(ctx);
	else	res=dv;
	if (sdiv == NULL || res == 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;

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

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

	/* 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--, resp--)
		{
		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);
#ifdef BN_DEBUG_LEVITTE
			TINYCLR_SSL_FPRINTF(OPENSSL_TYPE__FILE_STDERR,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
X) -> 0x%08X\n",
				n0, n1, d0, q);
#endif
#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);
#ifdef BN_DEBUG_LEVITTE
			TINYCLR_SSL_FPRINTF(OPENSSL_TYPE__FILE_STDERR,"DEBUG: bn_div_words(0x%08X,0x%08X,0x%08\
X) -> 0x%08X\n",
				n0, n1, d0, q);
#endif
#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 = 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);
		}
	bn_correct_top(res);
	BN_CTX_end(ctx);
	return(1);
err:
	bn_check_top(rm);
	BN_CTX_end(ctx);
	return(0);
	}
Exemple #3
0
/*-
 * 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;
    wnum.flags = BN_FLG_STATIC_DATA;
    /*
     * 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;
}