[PATCH] ARM: Fix VFP to use do_div()

VFP used __divdi3 64-bit division needlessly.  Convert it to use
our 64-bit by 32-bit division instead.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

arch/arm/vfp/vfp.h

@@ -117,7 +117,13 @@ static inline u64 vfp_estimate_div128to64(u64 nh, u64 nl, u64 m)
 	if (nh >= m)
 		return ~0ULL;
 	mh = m >> 32;
-	z = (mh << 32 <= nh) ? 0xffffffff00000000ULL : (nh / mh) << 32;
+	if (mh << 32 <= nh) {
+		z = 0xffffffff00000000ULL;
+	} else {
+		z = nh;
+		do_div(z, mh);
+		z <<= 32;
+	}
 	mul64to128(&termh, &terml, m, z);
 	sub128(&remh, &reml, nh, nl, termh, terml);
 	ml = m << 32;
@@ -126,7 +132,12 @@ static inline u64 vfp_estimate_div128to64(u64 nh, u64 nl, u64 m)
 		add128(&remh, &reml, remh, reml, mh, ml);
 	}
 	remh = (remh << 32) | (reml >> 32);
-	z |= (mh << 32 <= remh) ? 0xffffffff : remh / mh;
+	if (mh << 32 <= remh) {
+		z |= 0xffffffff;
+	} else {
+		do_div(remh, mh);
+		z |= remh;
+	}
 	return z;
 }
 

arch/arm/vfp/vfpdouble.c

@@ -32,6 +32,8 @@
  */
 #include <linux/kernel.h>
 #include <linux/bitops.h>
+
+#include <asm/div64.h>
 #include <asm/ptrace.h>
 #include <asm/vfp.h>
 

arch/arm/vfp/vfpsingle.c

@@ -32,6 +32,8 @@
  */
 #include <linux/kernel.h>
 #include <linux/bitops.h>
+
+#include <asm/div64.h>
 #include <asm/ptrace.h>
 #include <asm/vfp.h>
 
@@ -303,7 +305,11 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand)
 		if (z <= a)
 			return (s32)a >> 1;
 	}
-	return (u32)(((u64)a << 31) / z) + (z >> 1);
+	{
+		u64 v = (u64)a << 31;
+		do_div(v, z);
+		return v + (z >> 1);
+	}
 }
 
 static u32 vfp_single_fsqrt(int sd, int unused, s32 m, u32 fpscr)
@@ -1107,7 +1113,11 @@ static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr)
 		vsn.significand >>= 1;
 		vsd.exponent++;
 	}
-	vsd.significand = ((u64)vsn.significand << 32) / vsm.significand;
+	{
+		u64 significand = (u64)vsn.significand << 32;
+		do_div(significand, vsm.significand);
+		vsd.significand = significand;
+	}
 	if ((vsd.significand & 0x3f) == 0)
 		vsd.significand |= ((u64)vsm.significand * vsd.significand != (u64)vsn.significand << 32);