Merge pull request JuliaLang#14278 from JuliaLang/yyc/gc/cleanup

Do not assume the GC is always running on the master thread.

Author: yuyichao

src/gc.c

@@ -1028,7 +1028,7 @@ static size_t array_nbytes(jl_array_t *a)
     return sz;
 }
 
-void jl_gc_free_array(jl_array_t *a)
+static void jl_gc_free_array(jl_array_t *a)
 {
     if (a->how == 2) {
         char *d = (char*)a->data - a->offset*a->elsize;
@@ -1630,19 +1630,21 @@ static void gc_mark_stack(jl_value_t* ta, jl_gcframe_t *s, ptrint_t offset, int
 static void gc_mark_task_stack(jl_task_t *ta, int d)
 {
     int stkbuf = (ta->stkbuf != (void*)(intptr_t)-1 && ta->stkbuf != NULL);
+    int16_t tid = ta->tid;
+    jl_tls_states_t *ptls = jl_all_task_states[tid].ptls;
     if (stkbuf) {
 #ifndef COPY_STACKS
-        if (ta != jl_root_task) // stkbuf isn't owned by julia for the root task
+        if (ta != ptls->root_task) // stkbuf isn't owned by julia for the root task
 #endif
         gc_setmark_buf(ta->stkbuf, gc_bits(jl_astaggedvalue(ta)));
     }
-    if (ta == jl_all_task_states[ta->tid].ptls->current_task) {
-        gc_mark_stack((jl_value_t*)ta, *jl_all_pgcstacks[ta->tid], 0, d);
+    if (ta == ptls->current_task) {
+        gc_mark_stack((jl_value_t*)ta, ptls->pgcstack, 0, d);
     }
     else if (stkbuf) {
         ptrint_t offset;
 #ifdef COPY_STACKS
-        offset = (char *)ta->stkbuf - ((char *)jl_stackbase - ta->ssize);
+        offset = (char *)ta->stkbuf - ((char *)ptls->stackbase - ta->ssize);
 #else
         offset = 0;
 #endif
@@ -1907,8 +1909,6 @@ static void pre_mark(void)
 
     // invisible builtin values
     if (jl_an_empty_cell) gc_push_root(jl_an_empty_cell, 0);
-    gc_push_root(jl_exception_in_transit, 0);
-    gc_push_root(jl_task_arg_in_transit, 0);
     if (jl_module_init_order != NULL)
         gc_push_root(jl_module_init_order, 0);
 
@@ -2037,7 +2037,7 @@ static void print_obj_profile(htable_t nums, htable_t sizes)
     }
 }
 
-void print_obj_profiles(void)
+static void print_obj_profiles(void)
 {
     jl_printf(JL_STDERR, "Transient mark :\n");
     print_obj_profile(obj_counts[0], obj_sizes[0]);
@@ -2054,10 +2054,6 @@ static int saved_mark_sp = 0;
 static int sweep_mask = GC_MARKED;
 #define MIN_SCAN_BYTES 1024*1024
 
-void prepare_sweep(void)
-{
-}
-
 JL_DLLEXPORT void jl_gc_collect(int full)
 {
     if (!is_gc_enabled) return;
@@ -2466,8 +2462,9 @@ void jl_print_gc_stats(JL_STREAM *s)
 jl_thread_heap_t *jl_mk_thread_heap(void)
 {
 #ifdef JULIA_ENABLE_THREADING
-    // FIXME - should be cache-aligned malloc
-    jl_thread_heap = (jl_thread_heap_t*)malloc(sizeof(jl_thread_heap_t));
+    // Cache-aligned malloc
+    jl_thread_heap =
+        (jl_thread_heap_t*)jl_malloc_aligned(sizeof(jl_thread_heap_t), 64);
 #endif
     FOR_CURRENT_HEAP () {
         const int* szc = sizeclasses;

src/threading.c

@@ -127,7 +127,6 @@ JL_DLLEXPORT JL_CONST_FUNC jl_tls_states_t *(jl_get_ptls_states)(void)
 JL_DLLEXPORT int jl_n_threads;     // # threads we're actually using
 JL_DLLEXPORT int jl_max_threads;   // # threads possible
 jl_thread_task_state_t *jl_all_task_states;
-jl_gcframe_t ***jl_all_pgcstacks;
 
 // return calling thread's ID
 JL_DLLEXPORT int16_t jl_threadid(void) { return ti_tid; }
@@ -139,7 +138,6 @@ static void ti_initthread(int16_t tid)
 {
     ti_tid = tid;
     jl_pgcstack = NULL;
-    jl_all_pgcstacks[tid] = &jl_pgcstack;
 #ifdef JULIA_ENABLE_THREADING
     jl_all_heaps[tid] = jl_mk_thread_heap();
 #else
@@ -291,7 +289,6 @@ void jl_init_threading(void)
 
     // set up space for per-thread heaps
     jl_all_heaps = (struct _jl_thread_heap_t **)malloc(jl_n_threads * sizeof(void*));
-    jl_all_pgcstacks = (jl_gcframe_t ***)malloc(jl_n_threads * sizeof(void*));
     jl_all_task_states = (jl_thread_task_state_t *)malloc(jl_n_threads * sizeof(jl_thread_task_state_t));
 
 #if PROFILE_JL_THREADING
@@ -538,7 +535,6 @@ void jl_init_threading(void)
     jl_all_task_states = &_jl_all_task_states;
     jl_max_threads = 1;
     jl_n_threads = 1;
-    jl_all_pgcstacks = (jl_gcframe_t***) malloc(jl_n_threads * sizeof(jl_gcframe_t**));
 
 #if defined(__linux__) && defined(JL_USE_INTEL_JITEVENTS)
     if (jl_using_intel_jitevents)

src/threading.h

@@ -22,7 +22,6 @@ extern JL_DLLEXPORT int jl_n_threads;  // # threads we're actually using
 // GC
 extern struct _jl_thread_heap_t **jl_all_heaps;
 #endif
-extern jl_gcframe_t ***jl_all_pgcstacks;
 
 // thread state
 enum {