diff --git a/src/Makefile b/src/Makefile
index 52e673aa6cc1aa..7fcc2cfa2ffebb 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -318,6 +318,7 @@ $(BUILDDIR)/debuginfo.o $(BUILDDIR)/debuginfo.dbg.obj: $(addprefix $(SRCDIR)/,de
 $(BUILDDIR)/disasm.o $(BUILDDIR)/disasm.dbg.obj: $(SRCDIR)/debuginfo.h $(SRCDIR)/processor.h
 $(BUILDDIR)/gc-debug.o $(BUILDDIR)/gc-debug.dbg.obj: $(SRCDIR)/gc-common.h $(SRCDIR)/gc-stock.h
 $(BUILDDIR)/gc-pages.o $(BUILDDIR)/gc-pages.dbg.obj: $(SRCDIR)/gc-common.h $(SRCDIR)/gc-stock.h
+$(BUILDDIR)/gc-stacks.o $(BUILDDIR)/gc-stacks.dbg.obj: $(SRCDIR)/gc-common.h $(SRCDIR)/gc-stock.h
 $(BUILDDIR)/gc-stock.o $(BUILDDIR)/gc.dbg.obj: $(SRCDIR)/gc-common.h $(SRCDIR)/gc-stock.h $(SRCDIR)/gc-heap-snapshot.h $(SRCDIR)/gc-alloc-profiler.h $(SRCDIR)/gc-page-profiler.h
 $(BUILDDIR)/gc-heap-snapshot.o $(BUILDDIR)/gc-heap-snapshot.dbg.obj: $(SRCDIR)/gc-heap-snapshot.h
 $(BUILDDIR)/gc-alloc-profiler.o $(BUILDDIR)/gc-alloc-profiler.dbg.obj: $(SRCDIR)/gc-alloc-profiler.h
diff --git a/src/gc-stacks.c b/src/gc-stacks.c
index 0d17eafbcfd1e5..c7798dbb34a8b4 100644
--- a/src/gc-stacks.c
+++ b/src/gc-stacks.c
@@ -1,6 +1,7 @@
 // This file is a part of Julia. License is MIT: https://julialang.org/license
 
 #include "gc-common.h"
+#include "gc-stock.h"
 #include "threading.h"
 #ifndef _OS_WINDOWS_
 #  include <sys/resource.h>
@@ -202,11 +203,8 @@ JL_DLLEXPORT void *jl_malloc_stack(size_t *bufsz, jl_task_t *owner) JL_NOTSAFEPO
     return stk;
 }
 
-extern _Atomic(int) gc_ptls_sweep_idx;
-extern _Atomic(int) gc_n_threads_sweeping;
-extern _Atomic(int) gc_stack_free_idx;
 
-void sweep_stack_pools(void) JL_NOTSAFEPOINT
+void sweep_stack_pool_loop(void) JL_NOTSAFEPOINT
 {
     // Stack sweeping algorithm:
     //    // deallocate stacks if we have too many sitting around unused
diff --git a/src/gc-stock.c b/src/gc-stock.c
index ba0ef74515e8f6..00c936be5f601d 100644
--- a/src/gc-stock.c
+++ b/src/gc-stock.c
@@ -1027,6 +1027,20 @@ void gc_sweep_wait_for_all_stacks(void)
     }
 }
 
+void sweep_stack_pools(jl_ptls_t ptls) JL_NOTSAFEPOINT
+{
+    // initialize ptls index for parallel sweeping of stack pools
+    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);
@@ -3095,16 +3109,7 @@ static int _jl_gc_collect(jl_ptls_t ptls, jl_gc_collection_t collection)
 #endif
         current_sweep_full = sweep_full;
         sweep_weak_refs();
-        // initialize ptls index for parallel sweeping of stack pools
-        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_pools();
-        gc_sweep_wait_for_all_stacks();
+        sweep_stack_pools(ptls);
         gc_sweep_other(ptls, sweep_full);
         gc_scrub();
         gc_verify_tags();
@@ -3516,6 +3521,10 @@ STATIC_INLINE int may_sweep(jl_ptls_t ptls) JL_NOTSAFEPOINT
     return (jl_atomic_load(&ptls->gc_tls.gc_sweeps_requested) > 0);
 }
 
+STATIC_INLINE int may_sweep_stack(jl_ptls_t ptls) JL_NOTSAFEPOINT
+{
+    return (jl_atomic_load(&ptls->gc_tls.gc_stack_sweep_requested) > 0);
+}
 // parallel gc thread function
 void jl_parallel_gc_threadfun(void *arg)
 {
@@ -3544,10 +3553,14 @@ void jl_parallel_gc_threadfun(void *arg)
         uv_mutex_unlock(&gc_threads_lock);
         assert(jl_atomic_load_relaxed(&ptls->gc_state) == JL_GC_PARALLEL_COLLECTOR_THREAD);
         gc_mark_loop_parallel(ptls, 0);
+        if (may_sweep_stack(ptls)) {
+            assert(jl_atomic_load_relaxed(&ptls->gc_state) == JL_GC_PARALLEL_COLLECTOR_THREAD);
+            sweep_stack_pool_loop();
+            jl_atomic_fetch_add(&ptls->gc_tls.gc_stack_sweep_requested, -1);
+        }
         if (may_sweep(ptls)) {
             assert(jl_atomic_load_relaxed(&ptls->gc_state) == JL_GC_PARALLEL_COLLECTOR_THREAD);
             gc_sweep_pool_parallel(ptls);
-            sweep_stack_pools();
             jl_atomic_fetch_add(&ptls->gc_tls.gc_sweeps_requested, -1);
         }
     }
diff --git a/src/gc-stock.h b/src/gc-stock.h
index 45c93bf4289ae9..afed1d99c8d911 100644
--- a/src/gc-stock.h
+++ b/src/gc-stock.h
@@ -511,6 +511,8 @@ 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 _Atomic(int) n_threads_running;
 extern uv_barrier_t thread_init_done;
 void gc_mark_queue_all_roots(jl_ptls_t ptls, jl_gc_markqueue_t *mq);
@@ -521,7 +523,7 @@ 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) JL_NOTSAFEPOINT;
+void sweep_stack_pool_loop(void) JL_NOTSAFEPOINT;
 void jl_gc_debug_init(void);
 
 // GC pages
diff --git a/src/gc-tls.h b/src/gc-tls.h
index 9e4b09404db846..183016cd915151 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;