Merge branch 'master' into msbuilddll2

[mono.git] / mono / metadata / sgen-stw.c

diff --git a/mono/metadata/sgen-stw.c b/mono/metadata/sgen-stw.c
old mode 100644 (file)
new mode 100755 (executable)
index 92326a8..54dc218
--- a/mono/metadata/sgen-stw.c
+++ b/mono/metadata/sgen-stw.c
@@ -58,7 +58,7 @@ static void
 update_current_thread_stack (void *start)
 {
        int stack_guard = 0;
-#ifndef USE_MONO_CTX
+#if !defined(USE_MONO_CTX)
        void *reg_ptr = cur_thread_regs;
 #endif
        SgenThreadInfo *info = mono_thread_info_current ();
@@ -91,11 +91,17 @@ is_ip_in_managed_allocator (MonoDomain *domain, gpointer ip)
                return FALSE;
        if (!sgen_has_critical_method ())
                return FALSE;
-       ji = mono_jit_info_table_find (domain, ip);
+
+       /*
+        * mono_jit_info_table_find is not async safe since it calls into the AOT runtime to load information for
+        * missing methods (#13951). To work around this, we disable the AOT fallback. For this to work, the JIT needs
+        * to register the jit info for all GC critical methods after they are JITted/loaded.
+        */
+       ji = mono_jit_info_table_find_internal (domain, ip, FALSE);
        if (!ji)
                return FALSE;
 
-       return sgen_is_critical_method (ji->method);
+       return sgen_is_critical_method (mono_jit_info_get_method (ji));
 }
 
 static int
@@ -111,9 +117,9 @@ restart_threads_until_none_in_managed_allocator (void)
                   allocator */
                FOREACH_THREAD_SAFE (info) {
                        gboolean result;
-                       if (info->skip || info->gc_disabled || !info->joined_stw)
+                       if (info->skip || info->gc_disabled)
                                continue;
-                       if (!info->thread_is_dying && (!info->stack_start || info->in_critical_region ||
+                       if (mono_thread_info_run_state (info) == STATE_RUNNING && (!info->stack_start || info->in_critical_region || info->info.inside_critical_region ||
                                        is_ip_in_managed_allocator (info->stopped_domain, info->stopped_ip))) {
                                binary_protocol_thread_restart ((gpointer)mono_thread_info_get_tid (info));
                                SGEN_LOG (3, "thread %p resumed.", (void*) (size_t) info->info.native_handle);
@@ -189,18 +195,6 @@ release_gc_locks (void)
        UNLOCK_INTERRUPTION;
 }
 
-static void
-stw_bridge_process (void)
-{
-       sgen_bridge_processing_stw_step ();
-}
-
-static void
-bridge_process (int generation)
-{
-       sgen_bridge_processing_finish (generation);
-}
-
 static TV_DECLARE (stop_world_time);
 static unsigned long max_pause_usec = 0;
 
@@ -233,6 +227,7 @@ sgen_stop_world (int generation)
        MONO_GC_WORLD_STOP_END ();
 
        sgen_memgov_collection_start (generation);
+       sgen_bridge_reset_data ();
 
        return count;
 }
@@ -248,6 +243,7 @@ sgen_restart_world (int generation, GGTimingInfo *timing)
        unsigned long usec, bridge_usec;
 
        /* notify the profiler of the leftovers */
+       /* FIXME this is the wrong spot at we can STW for non collection reasons. */
        if (G_UNLIKELY (mono_profiler_events & MONO_PROFILE_GC_MOVES))
                sgen_gc_event_moves ();
        mono_profiler_gc_event (MONO_GC_EVENT_PRE_START_WORLD, generation);
@@ -261,9 +257,6 @@ sgen_restart_world (int generation, GGTimingInfo *timing)
 #endif
        } END_FOREACH_THREAD
 
-       stw_bridge_process ();
-       release_gc_locks ();
-
        count = sgen_thread_handshake (FALSE);
        TV_GETTIME (end_sw);
        usec = TV_ELAPSED (stop_world_time, end_sw);
@@ -272,9 +265,21 @@ sgen_restart_world (int generation, GGTimingInfo *timing)
        mono_profiler_gc_event (MONO_GC_EVENT_POST_START_WORLD, generation);
        MONO_GC_WORLD_RESTART_END (generation);
 
-       mono_thread_hazardous_try_free_some ();
+       /*
+        * We must release the thread info suspend lock after doing
+        * the thread handshake.  Otherwise, if the GC stops the world
+        * and a thread is in the process of starting up, but has not
+        * yet registered (it's not in the thread_list), it is
+        * possible that the thread does register while the world is
+        * stopped.  When restarting the GC will then try to restart
+        * said thread, but since it never got the suspend signal, it
+        * cannot answer the restart signal, so a deadlock results.
+        */
+       release_gc_locks ();
+
+       sgen_try_free_some_memory = TRUE;
 
-       bridge_process (generation);
+       sgen_bridge_processing_finish (generation);
 
        TV_GETTIME (end_bridge);
        bridge_usec = TV_ELAPSED (end_sw, end_bridge);