diff --git a/cranelift/codegen/meta/src/isa/x86/encodings.rs b/cranelift/codegen/meta/src/isa/x86/encodings.rs
index c5bbfe1a83..a71b2cc68f 100644
--- a/cranelift/codegen/meta/src/isa/x86/encodings.rs
+++ b/cranelift/codegen/meta/src/isa/x86/encodings.rs
@@ -683,10 +683,7 @@ pub fn define(
     for &ty in &[B1, B8, B16, B32] {
         e.enc_both(bconst.bind(ty), rec_pu_id_bool.opcodes(vec![0xb8]));
     }
-    e.enc64(
-        bconst.bind(B64),
-        rec_pu_id_bool.opcodes(vec![0xb8]).rex().w(),
-    );
+    e.enc64(bconst.bind(B64), rec_pu_id_bool.opcodes(vec![0xb8]).rex());
 
     // Shifts and rotates.
     // Note that the dynamic shift amount is only masked by 5 or 6 bits; the 8-bit
diff --git a/cranelift/filetests/filetests/isa/x86/binary64-run.clif b/cranelift/filetests/filetests/isa/x86/binary64-run.clif
new file mode 100644
index 0000000000..b255770c1e
--- /dev/null
+++ b/cranelift/filetests/filetests/isa/x86/binary64-run.clif
@@ -0,0 +1,10 @@
+test run
+target x86_64
+
+; this verifies that returning b64 immediates does not result in a segmentation fault, see https://github.com/CraneStation/cranelift/issues/911
+function %test_b64() -> b64 {
+ebb0:
+[-, %r10]   v0 = bconst.b64 true
+            return v0
+}
+; run
diff --git a/cranelift/filetests/filetests/isa/x86/binary64.clif b/cranelift/filetests/filetests/isa/x86/binary64.clif
index a65b3d3d1d..7742a24ee3 100644
--- a/cranelift/filetests/filetests/isa/x86/binary64.clif
+++ b/cranelift/filetests/filetests/isa/x86/binary64.clif
@@ -1642,3 +1642,10 @@ ebb0:
 
     return
 }
+
+function %B64() {
+ebb0:
+    [-, %rax]   v1 = bconst.b64 true    ; bin: 40 b8 00000001
+    [-, %r10]   v0 = bconst.b64 true    ; bin: 41 ba 00000001
+    return
+}