diff --git a/src/libcore/task.rs b/src/libcore/task.rs
index 28f9486a67793..85d1ee3da94cc 100644
--- a/src/libcore/task.rs
+++ b/src/libcore/task.rs
@@ -1473,9 +1473,13 @@ extern mod rustrt {
 
     fn rust_task_is_unwinding(task: *rust_task) -> bool;
     fn rust_osmain_sched_id() -> sched_id;
+    #[rust_stack]
     fn rust_task_inhibit_kill(t: *rust_task);
+    #[rust_stack]
     fn rust_task_allow_kill(t: *rust_task);
+    #[rust_stack]
     fn rust_task_inhibit_yield(t: *rust_task);
+    #[rust_stack]
     fn rust_task_allow_yield(t: *rust_task);
     fn rust_task_kill_other(task: *rust_task);
     fn rust_task_kill_all(task: *rust_task);
diff --git a/src/rt/rust_task.cpp b/src/rt/rust_task.cpp
index 061e87ebff82d..791c6a6551f97 100644
--- a/src/rt/rust_task.cpp
+++ b/src/rt/rust_task.cpp
@@ -631,27 +631,29 @@ rust_task::on_rust_stack() {
     }
 }
 
+// NB: In inhibit_kill and allow_kill, helgrind would complain that we need to
+// hold lifecycle_lock while accessing disallow_kill. Even though another
+// killing task may access disallow_kill concurrently, this is not racy
+// because the killer only cares if this task is blocking, and block() already
+// uses proper locking. See https://github.com/mozilla/rust/issues/3213 .
+
 void
 rust_task::inhibit_kill() {
-    scoped_lock with(lifecycle_lock);
     // Here might be good, though not mandatory, to check if we have to die.
     disallow_kill++;
 }
 
 void
 rust_task::allow_kill() {
-    scoped_lock with(lifecycle_lock);
     assert(disallow_kill > 0 && "Illegal allow_kill(): already killable!");
     disallow_kill--;
 }
 
 void rust_task::inhibit_yield() {
-    scoped_lock with(lifecycle_lock);
     disallow_yield++;
 }
 
 void rust_task::allow_yield() {
-    scoped_lock with(lifecycle_lock);
     assert(disallow_yield > 0 && "Illegal allow_yield(): already yieldable!");
     disallow_yield--;
 }