diff --git a/include/asm-powerpc/delay.h b/include/asm-powerpc/delay.h
index 057a60955474..f9200a65c632 100644
--- a/include/asm-powerpc/delay.h
+++ b/include/asm-powerpc/delay.h
@@ -17,5 +17,18 @@
 extern void __delay(unsigned long loops);
 extern void udelay(unsigned long usecs);
 
+/*
+ * On shared processor machines the generic implementation of mdelay can
+ * result in large errors. While each iteration of the loop inside mdelay
+ * is supposed to take 1ms, the hypervisor could sleep our partition for
+ * longer (eg 10ms). With the right timing these errors can add up.
+ *
+ * Since there is no 32bit overflow issue on 64bit kernels, just call
+ * udelay directly.
+ */
+#ifdef CONFIG_PPC64
+#define mdelay(n)	udelay((n) * 1000)
+#endif
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_POWERPC_DELAY_H */
diff --git a/include/linux/delay.h b/include/linux/delay.h
index acb74865b973..17ddb55430ae 100644
--- a/include/linux/delay.h
+++ b/include/linux/delay.h
@@ -25,10 +25,7 @@ extern unsigned long loops_per_jiffy;
 #define MAX_UDELAY_MS	5
 #endif
 
-#ifdef notdef
-#define mdelay(n) (\
-	{unsigned long __ms=(n); while (__ms--) udelay(1000);})
-#else
+#ifndef mdelay
 #define mdelay(n) (\
 	(__builtin_constant_p(n) && (n)<=MAX_UDELAY_MS) ? udelay((n)*1000) : \
 	({unsigned long __ms=(n); while (__ms--) udelay(1000);}))