diff --git a/src/gc-stacks.c b/src/gc-stacks.c
index 8515fb7797be6..998e763049164 100644
--- a/src/gc-stacks.c
+++ b/src/gc-stacks.c
@@ -101,6 +101,20 @@ JL_DLLEXPORT void jl_free_stack(void *stkbuf, size_t bufsz)
 }
 
 
+void jl_release_task_stack(jl_ptls_t ptls, jl_task_t *task)
+{
+    void *stkbuf = task->stkbuf;
+    size_t bufsz = task->bufsz;
+    if (bufsz <= pool_sizes[JL_N_STACK_POOLS - 1]) {
+        unsigned pool_id = select_pool(bufsz);
+        if (pool_sizes[pool_id] == bufsz) {
+            task->stkbuf = NULL;
+            arraylist_push(&ptls->heap.free_stacks[pool_id], stkbuf);
+        }
+    }
+}
+
+
 JL_DLLEXPORT void *jl_malloc_stack(size_t *bufsz, jl_task_t *owner)
 {
     jl_ptls_t ptls = jl_get_ptls_states();
diff --git a/src/task.c b/src/task.c
index 75656b82d0da7..04d6157b06d0b 100644
--- a/src/task.c
+++ b/src/task.c
@@ -191,6 +191,8 @@ JL_DLLEXPORT void julia_init(JL_IMAGE_SEARCH rel)
     _julia_init(rel);
 }
 
+void jl_release_task_stack(jl_ptls_t ptls, jl_task_t *task);
+
 static void ctx_switch(jl_ptls_t ptls, jl_task_t **pt)
 {
     jl_task_t *t = *pt;
@@ -223,10 +225,10 @@ static void ctx_switch(jl_ptls_t ptls, jl_task_t **pt)
     if (killed) {
         *pt = lastt; // can't fail after here: clear the gc-root for the target task now
         lastt->gcstack = NULL;
-        // if (!lastt->copy_stack) { // TODO: early free of stkbuf
-        //     jl_free_stack(lastt->stkbuf, lastt->bufsz);
-        //     lastt->stkbuf = NULL;
-        // }
+        if (!lastt->copy_stack && lastt->stkbuf) {
+            // early free of stkbuf back to the pool
+            jl_release_task_stack(ptls, lastt);
+        }
     }
     else {
 #ifdef COPY_STACKS