diff --git a/src/gc-stacks.c b/src/gc-stacks.c
index 1c1e2869607620..69e37db22c89fc 100644
--- a/src/gc-stacks.c
+++ b/src/gc-stacks.c
@@ -190,7 +190,7 @@ JL_DLLEXPORT void *jl_malloc_stack(size_t *bufsz, jl_task_t *owner) JL_NOTSAFEPO
     return stk;
 }
 
-void sweep_stack_pools(void)
+void sweep_stack_pool_loop(void)
 {
     // Stack sweeping algorithm:
     //    // deallocate stacks if we have too many sitting around unused
@@ -203,8 +203,11 @@ void sweep_stack_pools(void)
     //            bufsz = t->bufsz
     //            if (stkbuf)
     //                push(free_stacks[sz], stkbuf)
-    assert(gc_n_threads);
-    for (int i = 0; i < gc_n_threads; i++) {
+    jl_atomic_fetch_add(&gc_n_threads_sweeping, 1);
+    while (1) {
+        int i = jl_atomic_fetch_add_relaxed(&gc_ptls_sweep_idx, -1);
+        if (i < 0)
+            break;
         jl_ptls_t ptls2 = gc_all_tls_states[i];
 
         // free half of stacks that remain unused since last sweep
@@ -264,6 +267,7 @@ void sweep_stack_pools(void)
         }
         live_tasks->len -= ndel;
     }
+    jl_atomic_fetch_add(&gc_n_threads_sweeping, -1);
 }
 
 JL_DLLEXPORT jl_array_t *jl_live_tasks(void)
diff --git a/src/gc-tls.h b/src/gc-tls.h
index a9f711198e9142..6c9b814ca5661f 100644
--- a/src/gc-tls.h
+++ b/src/gc-tls.h
@@ -82,6 +82,7 @@ typedef struct {
     jl_gc_markqueue_t mark_queue;
     jl_gc_mark_cache_t gc_cache;
     _Atomic(size_t) gc_sweeps_requested;
+    _Atomic(uint8_t) gc_stack_sweep_requested;
     arraylist_t sweep_objs;
 } jl_gc_tls_states_t;
 
diff --git a/src/gc.c b/src/gc.c
index dad57687325450..9b174df9e15b68 100644
--- a/src/gc.c
+++ b/src/gc.c
@@ -26,6 +26,10 @@ _Atomic(int) gc_n_threads_sweeping;
 _Atomic(jl_gc_padded_page_stack_t *) gc_allocd_scratch;
 // `tid` of mutator thread that triggered GC
 _Atomic(int) gc_master_tid;
+// counter for sharing work when sweeping stacks
+_Atomic(int) gc_ptls_sweep_idx;
+// counter for round robin of giving back stack pages to the OS
+_Atomic(int) gc_stack_free_idx;
 // `tid` of first GC thread
 int gc_first_tid;
 // Mutex/cond used to synchronize wakeup of GC threads on parallel marking
@@ -1525,6 +1529,44 @@ static void gc_sweep_other(jl_ptls_t ptls, int sweep_full) JL_NOTSAFEPOINT
     gc_num.total_sweep_free_mallocd_memory_time += t_free_mallocd_memory_end - t_free_mallocd_memory_start;
 }
 
+// wake up all threads to sweep the stacks
+void gc_sweep_wake_all_stacks(jl_ptls_t ptls) JL_NOTSAFEPOINT
+{
+    uv_mutex_lock(&gc_threads_lock);
+    int first = gc_first_parallel_collector_thread_id();
+    int last = gc_last_parallel_collector_thread_id();
+    for (int i = first; i <= last; i++) {
+        jl_ptls_t ptls2 = gc_all_tls_states[i];
+        gc_check_ptls_of_parallel_collector_thread(ptls2);
+        jl_atomic_fetch_add(&ptls2->gc_tls.gc_stack_sweep_requested, 1);
+    }
+    uv_cond_broadcast(&gc_threads_cond);
+    uv_mutex_unlock(&gc_threads_lock);
+    return;
+}
+
+void gc_sweep_wait_for_all_stacks(void) JL_NOTSAFEPOINT
+{
+    while ((jl_atomic_load_acquire(&gc_ptls_sweep_idx)>= 0 ) || jl_atomic_load_acquire(&gc_n_threads_sweeping) != 0) {
+        jl_cpu_pause();
+    }
+}
+
+void sweep_stack_pools(jl_ptls_t ptls) JL_NOTSAFEPOINT
+{
+    // initialize ptls index for parallel sweeping of stack pools
+    assert(gc_n_threads);
+    int stack_free_idx = jl_atomic_load_relaxed(&gc_stack_free_idx);
+    if (stack_free_idx + 1 == gc_n_threads)
+        jl_atomic_store_relaxed(&gc_stack_free_idx, 0);
+    else
+        jl_atomic_store_relaxed(&gc_stack_free_idx, stack_free_idx + 1);
+    jl_atomic_store_release(&gc_ptls_sweep_idx, gc_n_threads - 1); // idx == gc_n_threads = release stacks to the OS so it's serial
+    gc_sweep_wake_all_stacks(ptls);
+    sweep_stack_pool_loop();
+    gc_sweep_wait_for_all_stacks();
+}
+
 static void gc_pool_sync_nfree(jl_gc_pagemeta_t *pg, jl_taggedvalue_t *last) JL_NOTSAFEPOINT
 {
     assert(pg->fl_begin_offset != UINT16_MAX);
@@ -3604,7 +3646,7 @@ static int _jl_gc_collect(jl_ptls_t ptls, jl_gc_collection_t collection)
 #endif
         current_sweep_full = sweep_full;
         sweep_weak_refs();
-        sweep_stack_pools();
+        sweep_stack_pools(ptls);
         gc_sweep_foreign_objs();
         gc_sweep_other(ptls, sweep_full);
         gc_scrub();
diff --git a/src/gc.h b/src/gc.h
index b06deec9d72389..e67fbeebc3d1f8 100644
--- a/src/gc.h
+++ b/src/gc.h
@@ -565,6 +565,8 @@ extern uv_cond_t gc_threads_cond;
 extern uv_sem_t gc_sweep_assists_needed;
 extern _Atomic(int) gc_n_threads_marking;
 extern _Atomic(int) gc_n_threads_sweeping;
+extern _Atomic(int) gc_ptls_sweep_idx;
+extern _Atomic(int) gc_stack_free_idx;
 extern uv_barrier_t thread_init_done;
 void gc_mark_queue_all_roots(jl_ptls_t ptls, jl_gc_markqueue_t *mq);
 void gc_mark_finlist_(jl_gc_markqueue_t *mq, jl_value_t *fl_parent, jl_value_t **fl_begin, jl_value_t **fl_end) JL_NOTSAFEPOINT;
@@ -574,7 +576,7 @@ void gc_mark_loop_serial(jl_ptls_t ptls);
 void gc_mark_loop_parallel(jl_ptls_t ptls, int master);
 void gc_sweep_pool_parallel(jl_ptls_t ptls);
 void gc_free_pages(void);
-void sweep_stack_pools(void);
+void sweep_stack_pool_loop(void);
 void jl_gc_debug_init(void);
 
 // GC pages