diff --git a/qemu/fpu/softfloat.c b/qemu/fpu/softfloat.c
index edece90b..23d93c74 100644
--- a/qemu/fpu/softfloat.c
+++ b/qemu/fpu/softfloat.c
@@ -1878,6 +1878,12 @@ static FloatParts scalbn_decomposed(FloatParts a, int n, float_status *s)
         return return_nan(a, s);
     }
     if (a.cls == float_class_normal) {
+        /* The largest float type (even though not supported by FloatParts)
+         * is float128, which has a 15 bit exponent.  Bounding N to 16 bits
+         * still allows rounding to infinity, without allowing overflow
+         * within the int32_t that backs FloatParts.exp.
+         */
+        n = MIN(MAX(n, -0x10000), 0x10000);
         a.exp += n;
     }
     return a;