[PATCH RFC: kvm tsc virtualization 16/20] Fix 32-bit mult_precise

[Zachary Amsden <zamsden@redhat.com>]

Turns out it wasn't very precise; it needs to work on 64-bit values.

Signed-off-by: Zachary Amsden <zamsden@redhat.com>
---
 arch/x86/kvm/x86.c |   18 +++++++++++++++---
 1 files changed, 15 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 7e2ba3e..792c895 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -803,15 +803,27 @@ static void compute_best_multiplier(unsigned long a, unsigned long b,
 	*s = shift;
 }
 
-static inline unsigned long mult_precise(unsigned long val, unsigned long mult,
-	int shift)
+static inline u64 mult_precise(u64 val, unsigned long mult, int shift)
 {
+#if (BITS_PER_LONG == 64)
 	unsigned long top, bot;
 
-	__asm__ ( "mul %3; shrd %1, %0" :
+	__asm__ ( "mulq %3; shrdq %1, %0" :
 		 "=&a" (bot), "=&d" (top) :
 		 "0" (mult), "rm" (val), "c" (shift));
 	return bot;
+#else
+	unsigned long ltop, lbot;
+	unsigned long htop, hbot;
+
+	__asm__ ( "mull %3; shrd %1, %0" :
+		 "=&a" (lbot), "=&d" (ltop) :
+		 "0" (mult), "rm" (long)(val), "c" (shift));
+	__asm__ ( "mull %3; shrd %1, %0" :
+		 "=&a" (hbot), "=&d" (htop) :
+		 "0" (mult), "rm" (long)(val >> 32), "c" (shift));
+	return (u64)lbot + ((u64)ltop + (u64)hbot) << 32;
+#endif
 }
 
 static inline u64 compute_ref_tsc(void)
-- 
1.6.5.2