diff --git a/src/rt/rust_builtin.cpp b/src/rt/rust_builtin.cpp
index 38e387735f977..7ffaccece6404 100644
--- a/src/rt/rust_builtin.cpp
+++ b/src/rt/rust_builtin.cpp
@@ -914,7 +914,19 @@ rust_signal_cond_lock(rust_cond_lock *lock) {
     }
 }
 
-
+// set/get/atexit task_local_data can run on the rust stack for speed.
+extern "C" void *
+rust_get_task_local_data(rust_task *task) {
+    return task->task_local_data;
+}
+extern "C" void
+rust_set_task_local_data(rust_task *task, void *data) {
+    task->task_local_data = data;
+}
+extern "C" void
+rust_task_local_data_atexit(rust_task *task, void (*cleanup_fn)(void *data)) {
+    task->task_local_data_cleanup = cleanup_fn;
+}
 
 //
 // Local Variables:
diff --git a/src/rt/rust_task.cpp b/src/rt/rust_task.cpp
index e5d61fda93d32..c899d9971970b 100644
--- a/src/rt/rust_task.cpp
+++ b/src/rt/rust_task.cpp
@@ -31,6 +31,8 @@ rust_task::rust_task(rust_sched_loop *sched_loop, rust_task_state state,
     propagate_failure(true),
     cc_counter(0),
     total_stack_sz(0),
+    task_local_data(NULL),
+    task_local_data_cleanup(NULL),
     state(state),
     cond(NULL),
     cond_name("none"),
@@ -115,6 +117,16 @@ cleanup_task(cleanup_args *args) {
         }
     }
 
+    // Clean up TLS. This will only be set if TLS was used to begin with.
+    // Because this is a crust function, it must be called from the C stack.
+    if (task->task_local_data_cleanup != NULL) {
+        // This assert should hold but it's not our job to ensure it (and
+        // the condition might change). Handled in libcore/task.rs.
+        // assert(task->task_local_data != NULL);
+        task->task_local_data_cleanup(task->task_local_data);
+        task->task_local_data = NULL;
+    }
+
     // FIXME (#2676): For performance we should do the annihilator
     // instead of the cycle collector even under normal termination, but
     // since that would hide memory management errors (like not derefing
diff --git a/src/rt/rust_task.h b/src/rt/rust_task.h
index e46b624c936d7..8e6a53892ec96 100644
--- a/src/rt/rust_task.h
+++ b/src/rt/rust_task.h
@@ -163,6 +163,10 @@ rust_task : public kernel_owned<rust_task>, rust_cond
     // The amount of stack we're using, excluding red zones
     size_t total_stack_sz;
 
+    // Used by rust task management routines in libcore/task.rs.
+    void *task_local_data;
+    void (*task_local_data_cleanup)(void *data);
+
 private:
 
     // Protects state, cond, cond_name
@@ -375,6 +379,10 @@ rust_task::call_on_c_stack(void *args, void *fn_ptr) {
     // Too expensive to check
     // assert(on_rust_stack());
 
+    // The shim functions generated by rustc contain the morestack prologue, so
+    // we need to let them know they have enough stack.
+    record_sp_limit(0);
+
     uintptr_t prev_rust_sp = next_rust_sp;
     next_rust_sp = get_sp();
 
@@ -398,12 +406,19 @@ rust_task::call_on_c_stack(void *args, void *fn_ptr) {
     }
 
     next_rust_sp = prev_rust_sp;
+
+    record_stack_limit();
 }
 
 inline void
 rust_task::call_on_rust_stack(void *args, void *fn_ptr) {
     // Too expensive to check
     // assert(!on_rust_stack());
+
+    // Because of the hack in the other function that disables the stack limit
+    // when entering the C stack, here we restore the stack limit again.
+    record_stack_limit();
+
     assert(get_sp_limit() != 0 && "Stack must be configured");
     assert(next_rust_sp);
 
@@ -427,6 +442,8 @@ rust_task::call_on_rust_stack(void *args, void *fn_ptr) {
         scoped_lock with(kill_lock);
         reentered_rust_stack = had_reentered_rust_stack;
     }
+
+    record_sp_limit(0);
 }
 
 inline void
diff --git a/src/rt/rust_upcall.cpp b/src/rt/rust_upcall.cpp
index 965d38fa70c93..9ce2258479324 100644
--- a/src/rt/rust_upcall.cpp
+++ b/src/rt/rust_upcall.cpp
@@ -49,19 +49,12 @@ extern "C" CDECL void
 upcall_call_shim_on_c_stack(void *args, void *fn_ptr) {
     rust_task *task = rust_get_current_task();
 
-    // FIXME (#1226) - The shim functions generated by rustc contain the
-    // morestack prologue, so we need to let them know they have enough
-    // stack.
-    record_sp_limit(0);
-
     try {
         task->call_on_c_stack(args, fn_ptr);
     } catch (...) {
         // Logging here is not reliable
         assert(false && "Foreign code threw an exception");
     }
-
-    task->record_stack_limit();
 }
 
 /*
@@ -72,11 +65,6 @@ extern "C" CDECL void
 upcall_call_shim_on_rust_stack(void *args, void *fn_ptr) {
     rust_task *task = rust_get_current_task();
 
-    // FIXME (#2680): Because of the hack in the other function that disables
-    // the stack limit when entering the C stack, here we restore the stack
-    // limit again.
-    task->record_stack_limit();
-
     try {
         task->call_on_rust_stack(args, fn_ptr);
     } catch (...) {
@@ -85,9 +73,6 @@ upcall_call_shim_on_rust_stack(void *args, void *fn_ptr) {
         // Logging here is not reliable
         assert(false && "Rust task failed after reentering the Rust stack");
     }
-
-    // FIXME (#2680): As above
-    record_sp_limit(0);
 }
 
 /**********************************************************************/
diff --git a/src/rt/rustrt.def.in b/src/rt/rustrt.def.in
index 9f702211c16f8..78ee8e4194497 100644
--- a/src/rt/rustrt.def.in
+++ b/src/rt/rustrt.def.in
@@ -167,3 +167,6 @@ rust_lock_cond_lock
 rust_unlock_cond_lock
 rust_wait_cond_lock
 rust_signal_cond_lock
+rust_get_task_local_data
+rust_set_task_local_data
+rust_task_local_data_atexit