Improve jl_print_task_backtraces()

We use `jl_rec_backtrace()` which tries to set the task's tid to the
current thread before gathering the backtrace. This will fail for
tasks that are sticky to another thread as their tid is never reset.
However, for `jl_print_task_backtraces()`, we aren't concerned about
thread safety since we assume that all threads are stopped so we add
a flag to `jl_rec_backtrace()` to ignore the task's tid.

With this, `jl_print_task_backtraces()` should now only miss tasks
that are currently executing on threads other than the calling thread.

src/stackwalk.c

@@ -856,7 +856,7 @@ _os_ptr_munge(uintptr_t ptr)
 
 extern bt_context_t *jl_to_bt_context(void *sigctx);
 
-void jl_rec_backtrace(jl_task_t *t) JL_NOTSAFEPOINT
+void jl_rec_backtrace(jl_task_t *t, int ignore_tid) JL_NOTSAFEPOINT
 {
     jl_task_t *ct = jl_current_task;
     jl_ptls_t ptls = ct->ptls;
@@ -868,8 +868,9 @@ void jl_rec_backtrace(jl_task_t *t) JL_NOTSAFEPOINT
     if (t->copy_stack || !t->started || t->stkbuf == NULL)
         return;
     int16_t old = -1;
-    if (!jl_atomic_cmpswap(&t->tid, &old, ptls->tid) && old != ptls->tid)
-        return;
+    if (!ignore_tid)
+        if (!jl_atomic_cmpswap(&t->tid, &old, ptls->tid) && old != ptls->tid)
+            return;
     bt_context_t *context = NULL;
 #if defined(_OS_WINDOWS_)
     bt_context_t c;
@@ -1079,10 +1080,18 @@ void jl_rec_backtrace(jl_task_t *t) JL_NOTSAFEPOINT
 #endif
     if (context)
         ptls->bt_size = rec_backtrace_ctx(ptls->bt_data, JL_MAX_BT_SIZE, context, t->gcstack);
-    if (old == -1)
+    if (!ignore_tid && old == -1)
         jl_atomic_store_relaxed(&t->tid, old);
 }
 
+void print_bt(size_t bt_size, jl_bt_element_t *bt_data)
+{
+    size_t i;
+    for (i = 0; i < bt_size; i += jl_bt_entry_size(bt_data + i)) {
+        jl_print_bt_entry_codeloc(bt_data + i);
+    }
+}
+
 //--------------------------------------------------
 // Tools for interactive debugging in gdb
 
@@ -1100,24 +1109,16 @@ JL_DLLEXPORT void jlbacktrace(void) JL_NOTSAFEPOINT
     jl_excstack_t *s = ct->excstack;
     if (!s)
         return;
-    size_t i, bt_size = jl_excstack_bt_size(s, s->top);
-    jl_bt_element_t *bt_data = jl_excstack_bt_data(s, s->top);
-    for (i = 0; i < bt_size; i += jl_bt_entry_size(bt_data + i)) {
-        jl_print_bt_entry_codeloc(bt_data + i);
-    }
+    print_bt(jl_excstack_bt_size(s, s->top), jl_excstack_bt_data(s, s->top));
 }
 
 // Print backtrace for specified task
 JL_DLLEXPORT void jlbacktracet(jl_task_t *t) JL_NOTSAFEPOINT
 {
     jl_task_t *ct = jl_current_task;
     jl_ptls_t ptls = ct->ptls;
-    jl_rec_backtrace(t);
-    size_t i, bt_size = ptls->bt_size;
-    jl_bt_element_t *bt_data = ptls->bt_data;
-    for (i = 0; i < bt_size; i += jl_bt_entry_size(bt_data + i)) {
-        jl_print_bt_entry_codeloc(bt_data + i);
-    }
+    jl_rec_backtrace(t, 0);
+    print_bt(ptls->bt_size, ptls->bt_data);
 }
 
 JL_DLLEXPORT void jl_print_backtrace(void) JL_NOTSAFEPOINT
@@ -1130,6 +1131,8 @@ JL_DLLEXPORT void jl_print_backtrace(void) JL_NOTSAFEPOINT
 // all of Julia's threads are not stopped!
 JL_DLLEXPORT void jl_print_task_backtraces(int show_done) JL_NOTSAFEPOINT
 {
+    jl_task_t *ct = jl_current_task;
+    jl_ptls_t ptls = ct->ptls;
     size_t nthreads = jl_atomic_load_acquire(&jl_n_threads);
     jl_ptls_t *allstates = jl_atomic_load_relaxed(&jl_all_tls_states);
     for (size_t i = 0; i < nthreads; i++) {
@@ -1143,7 +1146,8 @@ JL_DLLEXPORT void jl_print_task_backtraces(int show_done) JL_NOTSAFEPOINT
                 ptls2->root_task->sticky, ptls2->root_task->started,
                 jl_atomic_load_relaxed(&ptls2->root_task->_state),
                 jl_atomic_load_relaxed(&ptls2->root_task->tid) + 1);
-        jlbacktracet(ptls2->root_task);
+        jl_rec_backtrace(ptls2->root_task, 1);
+        print_bt(ptls->bt_size, ptls->bt_data);
 
         void **lst = live_tasks->items;
         for (size_t j = 0; j < live_tasks->len; j++) {
@@ -1156,10 +1160,16 @@ JL_DLLEXPORT void jl_print_task_backtraces(int show_done) JL_NOTSAFEPOINT
             jl_safe_printf("          (sticky: %d, started: %d, state: %d, tid: %d)\n",
                     t->sticky, t->started, t_state,
                     jl_atomic_load_relaxed(&t->tid) + 1);
-            if (t->stkbuf != NULL)
-                jlbacktracet(t);
-            else
+            if (t == ptls2->current_task) {
+                jl_safe_printf("      running\n");
+            }
+            else if (t->stkbuf == NULL) {
                 jl_safe_printf("      no stack\n");
+            }
+            else {
+                jl_rec_backtrace(t, 1);
+                print_bt(ptls->bt_size, ptls->bt_data);
+            }
             jl_safe_printf("     ---- End task %zu\n", j + 1);
         }
         jl_safe_printf("==== End thread %d\n", ptls2->tid + 1);