From 5f234f0fb5d047014b82232c744f0fd5c7ddcef9 Mon Sep 17 00:00:00 2001
From: Diogo Netto <61364108+d-netto@users.noreply.github.com>
Date: Tue, 16 Jul 2024 19:00:06 -0300
Subject: [PATCH] create separate function to spawn GC threads (#55108)

Third-party GCs (e.g. MMTk) will probably have their own function to
spawn GC threads.
---
 src/gc.c             | 20 ++++++++++++++++++++
 src/gc.h             |  1 +
 src/init.c           |  2 ++
 src/julia_internal.h |  2 ++
 src/threading.c      | 26 ++++++++------------------
 src/threading.h      |  2 ++
 6 files changed, 35 insertions(+), 18 deletions(-)

diff --git a/src/gc.c b/src/gc.c
index c70eee6905ee9..15d48f600ed80 100644
--- a/src/gc.c
+++ b/src/gc.c
@@ -3881,6 +3881,26 @@ void jl_init_thread_heap(jl_ptls_t ptls)
     jl_atomic_store_relaxed(&ptls->gc_tls.gc_num.allocd, -(int64_t)gc_num.interval);
 }
 
+void jl_start_gc_threads(void)
+{
+    int nthreads = jl_atomic_load_relaxed(&jl_n_threads);
+    int ngcthreads = jl_n_gcthreads;
+    int nmutator_threads = nthreads - ngcthreads;
+    uv_thread_t uvtid;
+    for (int i = nmutator_threads; i < nthreads; ++i) {
+        jl_threadarg_t *t = (jl_threadarg_t *)malloc_s(sizeof(jl_threadarg_t)); // ownership will be passed to the thread
+        t->tid = i;
+        t->barrier = &thread_init_done;
+        if (i == nthreads - 1 && jl_n_sweepthreads == 1) {
+            uv_thread_create(&uvtid, jl_concurrent_gc_threadfun, t);
+        }
+        else {
+            uv_thread_create(&uvtid, jl_parallel_gc_threadfun, t);
+        }
+        uv_thread_detach(&uvtid);
+    }
+}
+
 // System-wide initializations
 void jl_gc_init(void)
 {
diff --git a/src/gc.h b/src/gc.h
index 2ec249d322d94..b4d421c708547 100644
--- a/src/gc.h
+++ b/src/gc.h
@@ -562,6 +562,7 @@ 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 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;
 void gc_mark_finlist(jl_gc_markqueue_t *mq, arraylist_t *list, size_t start) JL_NOTSAFEPOINT;
diff --git a/src/init.c b/src/init.c
index 83f528b35ac11..0a0878a081e28 100644
--- a/src/init.c
+++ b/src/init.c
@@ -899,6 +899,8 @@ static NOINLINE void _finish_julia_init(JL_IMAGE_SEARCH rel, jl_ptls_t ptls, jl_
         jl_n_threads_per_pool[1] = 0;
     }
     jl_start_threads();
+    jl_start_gc_threads();
+    uv_barrier_wait(&thread_init_done);
 
     jl_init_heartbeat();
 
diff --git a/src/julia_internal.h b/src/julia_internal.h
index 75ebad73a4546..9ced95a8f027a 100644
--- a/src/julia_internal.h
+++ b/src/julia_internal.h
@@ -942,8 +942,10 @@ extern JL_DLLEXPORT ssize_t jl_tls_offset;
 extern JL_DLLEXPORT const int jl_tls_elf_support;
 void jl_init_threading(void);
 void jl_start_threads(void);
+void jl_start_gc_threads(void);
 
 // Whether the GC is running
+extern uv_mutex_t safepoint_lock;
 extern char *jl_safepoint_pages;
 STATIC_INLINE int jl_addr_is_safepoint(uintptr_t addr)
 {
diff --git a/src/threading.c b/src/threading.c
index 05607c12085f2..8f350d41f64b1 100644
--- a/src/threading.c
+++ b/src/threading.c
@@ -712,7 +712,7 @@ void jl_init_threading(void)
     gc_first_tid = nthreads + nthreadsi;
 }
 
-static uv_barrier_t thread_init_done;
+uv_barrier_t thread_init_done;
 
 void jl_start_threads(void)
 {
@@ -751,30 +751,20 @@ void jl_start_threads(void)
     uv_barrier_init(&thread_init_done, nthreads);
 
     // GC/System threads need to be after the worker threads.
-    int nworker_threads = nthreads - ngcthreads;
+    int nmutator_threads = nthreads - ngcthreads;
 
-    for (i = 1; i < nthreads; ++i) {
+    for (i = 1; i < nmutator_threads; ++i) {
         jl_threadarg_t *t = (jl_threadarg_t *)malloc_s(sizeof(jl_threadarg_t)); // ownership will be passed to the thread
         t->tid = i;
         t->barrier = &thread_init_done;
-        if (i < nworker_threads) {
-            uv_thread_create(&uvtid, jl_threadfun, t);
-            if (exclusive) {
-                mask[i] = 1;
-                uv_thread_setaffinity(&uvtid, mask, NULL, cpumasksize);
-                mask[i] = 0;
-            }
-        }
-        else if (i == nthreads - 1 && jl_n_sweepthreads == 1) {
-            uv_thread_create(&uvtid, jl_concurrent_gc_threadfun, t);
-        }
-        else {
-            uv_thread_create(&uvtid, jl_parallel_gc_threadfun, t);
+        uv_thread_create(&uvtid, jl_threadfun, t);
+        if (exclusive) {
+            mask[i] = 1;
+            uv_thread_setaffinity(&uvtid, mask, NULL, cpumasksize);
+            mask[i] = 0;
         }
         uv_thread_detach(&uvtid);
     }
-
-    uv_barrier_wait(&thread_init_done);
 }
 
 _Atomic(unsigned) _threadedregion; // HACK: keep track of whether to prioritize IO or threading
diff --git a/src/threading.h b/src/threading.h
index 260ecffa30dd5..cb26537699713 100644
--- a/src/threading.h
+++ b/src/threading.h
@@ -12,6 +12,8 @@ extern "C" {
 
 #define PROFILE_JL_THREADING            0
 
+extern uv_barrier_t thread_init_done;
+
 extern _Atomic(jl_ptls_t*) jl_all_tls_states JL_GLOBALLY_ROOTED; /* thread local storage */
 
 typedef struct _jl_threadarg_t {