diff --git a/arch/x86/core/intel64/thread.c b/arch/x86/core/intel64/thread.c
index 451f34b50ca..d0e44503948 100644
--- a/arch/x86/core/intel64/thread.c
+++ b/arch/x86/core/intel64/thread.c
@@ -12,6 +12,14 @@
 
 extern void x86_sse_init(struct k_thread *); /* in locore.S */
 
+/* FIXME: This exists to make space for a "return address" at the top
+ * of the stack.  Obviously this is unused at runtime, but is required
+ * for alignment: stacks at runtime should be 16-byte aligned, and a
+ * CALL will therefore push a return address that leaves the stack
+ * misaligned.  Effectively we're wasting 8 bytes here to undo (!) the
+ * alignment that the upper level code already tried to do for us.  We
+ * should clean this up.
+ */
 struct x86_initial_frame {
 	/* zeroed return address for ABI */
 	uint64_t rip;
diff --git a/arch/x86/core/userspace.c b/arch/x86/core/userspace.c
index 98027ff50fa..43d02af23ec 100644
--- a/arch/x86/core/userspace.c
+++ b/arch/x86/core/userspace.c
@@ -87,7 +87,7 @@ void *z_x86_userspace_prepare_thread(struct k_thread *thread)
 FUNC_NORETURN void arch_user_mode_enter(k_thread_entry_t user_entry,
 					void *p1, void *p2, void *p3)
 {
-	uint32_t stack_end;
+	size_t stack_end;
 
 	/* Transition will reset stack pointer to initial, discarding
 	 * any old context since this is a one-way operation
@@ -96,6 +96,15 @@ FUNC_NORETURN void arch_user_mode_enter(k_thread_entry_t user_entry,
 				      _current->stack_info.size -
 				      _current->stack_info.delta);
 
+#ifdef CONFIG_X86_64
+	/* x86_64 SysV ABI requires 16 byte stack alignment, which
+	 * means that on entry to a C function (which follows a CALL
+	 * that pushes 8 bytes) the stack must be MISALIGNED by
+	 * exactly 8 bytes.
+	 */
+	stack_end -= 8;
+#endif
+
 	z_x86_userspace_enter(user_entry, p1, p2, p3, stack_end,
 			      _current->stack_info.start);
 	CODE_UNREACHABLE;