Add CoreCLR support for android GC bridge (#116310)

* Add support for the new cross reference GCHandle type

From Aaron's implementation.

* Add interop glue for bridge

From Aaron's implementation

* Add GCBridge test

* Implement scanning of bridge objects and construction of SCCs

from Maoni

* Add tarjan bridge computation

* Fix object validation

Checking if the object is promoted was validating the next object header in debug builds. During bridge tarjan computation, we patch the object header for some objects in order to store data used by the bridge algorithm, so we need to disable this validation.

* bgc and multiple heap support for bridge obj processing

* Add incorrect waiting for gc bridge to finish

* Update code to new API

* Clear weakrefs for bridge objects with collected java peers

* Wait for bridge processing while resolving the gchandle to prevent races

* Maoni review

HANDLE_MAX_INTERNAL_TYPES value

new instead of malloc

assert for allocation failure

Reuse memory for ColorData and ScanData between collections. We still do alloc/free for other type of data, for example for arrays representing edges between SCCs.

Actually print class name when enabling tarjan bridge logs.

Add separate IsPromoted method to the gc interface

Rename TriggerGCBridge to TriggerClientBridgeProcessing to be more specific about what it is doing.

* Aaron review

* Update to new API

* Fix build

Include classes that are part of the Java Marshal API on mono as well

Disable test on mono and nativeaot.

* Use cross refs args directly in more places

* Change unreachable handle length from int to size_t

* Remove FEATURE_GCBRIDGE

* Move interop javamarshal code out of interop shared file

Create reset event during javamarshal initialize

* Use asserts in test

* Add bridge waiting to WeakReference<T> as well

* Review

* review

* High precision timestamp

* Naming consistency in gcbridge.cpp, matching rest of the runtime

* Fix interface breakage

* Fix test/JavaMarshal on platforms where it is not enabled

The test used to fail and the JavaMarshal api would crash due to missing internal calls.

* Share GetHighPrecisionTimeStamp across GC code

* Fix mark function used during BGC finishing pause

Author: BrzVlad

src/coreclr/System.Private.CoreLib/src/System/Runtime/InteropServices/GCHandle.CoreCLR.cs

@@ -41,6 +41,26 @@ private static void InternalFreeWithGCTransition(IntPtr dependentHandle)
         [LibraryImport(RuntimeHelpers.QCall, EntryPoint = "GCHandle_InternalFreeWithGCTransition")]
         private static partial void _InternalFreeWithGCTransition(IntPtr dependentHandle);
 
+#if FEATURE_JAVAMARSHAL
+        internal static object? InternalGetBridgeWait(IntPtr handle)
+        {
+            object? target = null;
+
+            if (GCHandle.InternalTryGetBridgeWait(handle, ref target))
+                return target;
+
+            InternalGetBridgeWait(handle, ObjectHandleOnStack.Create(ref target));
+
+            return target;
+        }
+
+        [MethodImpl(MethodImplOptions.InternalCall)]
+        private static extern bool InternalTryGetBridgeWait(IntPtr handle, ref object? result);
+
+        [LibraryImport(RuntimeHelpers.QCall, EntryPoint = "GCHandle_InternalGetBridgeWait")]
+        private static partial void InternalGetBridgeWait(IntPtr handle, ObjectHandleOnStack result);
+
+#endif
 #if DEBUG
         // The runtime performs additional checks in debug builds
         [MethodImpl(MethodImplOptions.InternalCall)]

src/coreclr/clr.featuredefines.props

@@ -6,6 +6,10 @@
         <ProfilingSupportedBuild>true</ProfilingSupportedBuild>
     </PropertyGroup>
 
+    <PropertyGroup Condition="'$(TargetsAndroid)' == 'true' OR '$(Configuration)' == 'debug' OR '$(Configuration)' == 'checked'">
+        <FeatureJavaMarshal>true</FeatureJavaMarshal>
+    </PropertyGroup>
+
     <PropertyGroup Condition="'$(TargetsUnix)' == 'true'">
         <FeatureXplatEventSource Condition="'$(FeatureXplatEventSource)' == '' AND '$(TargetOS)' == 'linux'">true</FeatureXplatEventSource>
         <FeatureComWrappers>true</FeatureComWrappers>
@@ -28,6 +32,7 @@
     </PropertyGroup>
 
     <PropertyGroup>
+        <DefineConstants Condition="'$(FeatureJavaMarshal)' == 'true'">$(DefineConstants);FEATURE_JAVAMARSHAL</DefineConstants>
         <DefineConstants Condition="'$(FeatureComWrappers)' == 'true'">$(DefineConstants);FEATURE_COMWRAPPERS</DefineConstants>
         <DefineConstants Condition="'$(FeatureCominterop)' == 'true'">$(DefineConstants);FEATURE_COMINTEROP</DefineConstants>
         <DefineConstants Condition="'$(FeatureCominteropApartmentSupport)' == 'true'">$(DefineConstants);FEATURE_COMINTEROP_APARTMENT_SUPPORT</DefineConstants>

src/coreclr/clrdefinitions.cmake

@@ -154,6 +154,8 @@ endif()
 
 # add_compile_definitions(FEATURE_RUNTIME_ASYNC)
 
+add_compile_definitions($<${FEATURE_JAVAMARSHAL}:FEATURE_JAVAMARSHAL>)
+
 add_compile_definitions($<$<NOT:$<BOOL:$<TARGET_PROPERTY:DAC_COMPONENT>>>:FEATURE_PROFAPI_ATTACH_DETACH>)
 
 add_definitions(-DFEATURE_READYTORUN)

src/coreclr/clrfeatures.cmake

@@ -62,6 +62,14 @@ if (CLR_CMAKE_TARGET_APPLE)
   set(FEATURE_OBJCMARSHAL 1)
 endif()
 
+if(NOT DEFINED FEATURE_JAVAMARSHAL)
+  if(CLR_CMAKE_TARGET_ANDROID)
+    set(FEATURE_JAVAMARSHAL 1)
+  else()
+    set(FEATURE_JAVAMARSHAL $<IF:$<CONFIG:Debug,Checked>,1,0>)
+  endif()
+endif()
+
 if (CLR_CMAKE_TARGET_WIN32)
   set(FEATURE_TYPEEQUIVALENCE 1)
 endif(CLR_CMAKE_TARGET_WIN32)

src/coreclr/debug/daccess/daccess.cpp

@@ -7700,11 +7700,18 @@ void CALLBACK DacHandleWalker::EnumCallback(PTR_UNCHECKED_OBJECTREF handle, uint
     data.Handle = TO_CDADDR(handle.GetAddr());
     data.Type = param->Type;
     if (param->Type == HNDTYPE_DEPENDENT)
+    {
         data.Secondary = GetDependentHandleSecondary(handle.GetAddr()).GetAddr();
-    else if (param->Type == HNDTYPE_WEAK_INTERIOR_POINTER)
+    }
+    else if (param->Type == HNDTYPE_WEAK_INTERIOR_POINTER
+        || param->Type == HNDTYPE_CROSSREFERENCE)
+    {
         data.Secondary = TO_CDADDR(HndGetHandleExtraInfo(handle.GetAddr()));
+    }
     else
+    {
         data.Secondary = 0;
+    }
     data.AppDomain = param->AppDomain;
     GetRefCountedHandleInfo((OBJECTREF)*handle, param->Type, &data.RefCount, &data.JupiterRefCount, &data.IsPegged, &data.StrongReference);
     data.StrongReference |= (BOOL)IsAlwaysStrongReference(param->Type);

src/coreclr/debug/daccess/request.cpp

@@ -3347,6 +3347,9 @@ HRESULT ClrDataAccess::GetHandleEnum(ISOSHandleEnum **ppHandleEnum)
 #if defined(FEATURE_COMINTEROP) || defined(FEATURE_COMWRAPPERS) || defined(FEATURE_OBJCMARSHAL)
                             HNDTYPE_REFCOUNTED,
 #endif // FEATURE_COMINTEROP || FEATURE_COMWRAPPERS || FEATURE_OBJCMARSHAL
+#if defined(FEATURE_JAVAMARSHAL)
+                            HNDTYPE_CROSSREFERENCE,
+#endif // FEATURE_JAVAMARSHAL
                             };
 
     return GetHandleEnumForTypes(types, ARRAY_SIZE(types), ppHandleEnum);

src/coreclr/gc/CMakeLists.txt

@@ -22,6 +22,7 @@ set(GC_SOURCES
   gceesvr.cpp
   gceewks.cpp
   gcload.cpp
+  gcbridge.cpp
   handletablecache.cpp)
 
 if(CLR_CMAKE_HOST_UNIX)
@@ -55,6 +56,7 @@ if (CLR_CMAKE_TARGET_WIN32)
     env/volatile.h
     gc.h
     gcconfig.h
+    gcbridge.h
     gcdesc.h
     gcenv.ee.standalone.inl
     gcenv.inl

src/coreclr/gc/env/gcenv.ee.h

@@ -39,6 +39,8 @@ class GCToEEInterface
     // Promote refcounted handle callback
     static bool RefCountedHandleCallbacks(Object * pObject);
 
+    static void TriggerClientBridgeProcessing(MarkCrossReferencesArgs* args);
+
     // Sync block cache management
     static void SyncBlockCacheWeakPtrScan(HANDLESCANPROC scanProc, uintptr_t lp1, uintptr_t lp2);
     static void SyncBlockCacheDemote(int max_gen);

src/coreclr/gc/env/gctoeeinterface.standalone.inl

@@ -49,6 +49,11 @@ namespace standalone
             return ::GCToEEInterface::RefCountedHandleCallbacks(pObject);
         }
 
+        void TriggerClientBridgeProcessing(MarkCrossReferencesArgs* args)
+        {
+            return ::GCToEEInterface::TriggerClientBridgeProcessing(args);
+        }
+
         void SyncBlockCacheWeakPtrScan(HANDLESCANPROC scanProc, uintptr_t lp1, uintptr_t lp2)
         {
             ::GCToEEInterface::SyncBlockCacheWeakPtrScan(scanProc, lp1, lp2);

src/coreclr/gc/gc.cpp

@@ -285,13 +285,6 @@ uint64_t qpf;
 double qpf_ms;
 double qpf_us;
 
-uint64_t GetHighPrecisionTimeStamp()
-{
-    int64_t ts = GCToOSInterface::QueryPerformanceCounter();
-
-    return (uint64_t)((double)ts * qpf_us);
-}
-
 uint64_t RawGetHighPrecisionTimeStamp()
 {
     return (uint64_t)GCToOSInterface::QueryPerformanceCounter();
@@ -670,7 +663,8 @@ enum gc_join_stage
     // No longer in use but do not remove, see comments for this enum.
     gc_join_disable_software_write_watch = 38,
     gc_join_merge_temp_fl = 39,
-    gc_join_max = 40
+    gc_join_bridge_processing = 40,
+    gc_join_max = 41
 };
 
 enum gc_join_flavor
@@ -2865,6 +2859,11 @@ FinalizerWorkItem* gc_heap::finalizer_work;
 BOOL gc_heap::proceed_with_gc_p = FALSE;
 GCSpinLock gc_heap::gc_lock;
 
+#ifdef FEATURE_JAVAMARSHAL
+uint8_t**   gc_heap::global_bridge_list;
+size_t      gc_heap::num_global_bridge_objs;
+#endif //FEATURE_JAVAMARSHAL
+
 #ifdef BACKGROUND_GC
 uint64_t gc_heap::total_uoh_a_last_bgc = 0;
 #endif //BACKGROUND_GC
@@ -30449,6 +30448,41 @@ void gc_heap::mark_phase (int condemned_gen_number)
     mark_queue.verify_empty();
     fire_mark_event (ETW::GC_ROOT_DH_HANDLES, current_promoted_bytes, last_promoted_bytes);
 
+#ifdef FEATURE_JAVAMARSHAL
+
+#ifdef MULTIPLE_HEAPS
+    dprintf(3, ("Joining for short weak handle scan"));
+    gc_t_join.join(this, gc_join_bridge_processing);
+    if (gc_t_join.joined())
+    {
+#endif //MULTIPLE_HEAPS
+        global_bridge_list = GCScan::GcProcessBridgeObjects (condemned_gen_number, max_generation, &sc, &num_global_bridge_objs);
+
+#ifdef MULTIPLE_HEAPS
+        dprintf (3, ("Starting all gc thread after bridge processing"));
+        gc_t_join.restart();
+    }
+#endif //MULTIPLE_HEAPS
+
+    {
+        int thread = heap_number;
+        // Each thread will receive an equal chunk of bridge objects, with the last thread
+        // handling a few more objects from the remainder.
+        size_t count_per_heap = num_global_bridge_objs / n_heaps;
+        size_t start_index = thread * count_per_heap;
+        size_t end_index = (thread == n_heaps - 1) ? num_global_bridge_objs : (thread + 1) * count_per_heap;
+
+        for (size_t obj_idx = start_index; obj_idx < end_index; obj_idx++)
+        {
+            mark_object_simple (&global_bridge_list[obj_idx] THREAD_NUMBER_ARG);
+        }
+
+        drain_mark_queue();
+        // using GC_ROOT_DH_HANDLES temporarily. add a new value for GC_ROOT_BRIDGE
+        fire_mark_event (ETW::GC_ROOT_DH_HANDLES, current_promoted_bytes, last_promoted_bytes);
+    }
+#endif //FEATURE_JAVAMARSHAL
+
 #ifdef MULTIPLE_HEAPS
     dprintf(3, ("Joining for short weak handle scan"));
     gc_t_join.join(this, gc_join_null_dead_short_weak);
@@ -39180,6 +39214,41 @@ void gc_heap::background_mark_phase ()
     dprintf (2, ("after NR 1st Hov count: %zu", bgc_overflow_count));
     bgc_overflow_count = 0;
 
+#ifdef FEATURE_JAVAMARSHAL
+
+    // FIXME Any reason this code should be different for BGC ? Otherwise extract it to some common method ?
+
+#ifdef MULTIPLE_HEAPS
+    dprintf(3, ("Joining for short weak handle scan"));
+    gc_t_join.join(this, gc_join_bridge_processing);
+    if (gc_t_join.joined())
+    {
+#endif //MULTIPLE_HEAPS
+        global_bridge_list = GCScan::GcProcessBridgeObjects (max_generation, max_generation, &sc, &num_global_bridge_objs);
+
+#ifdef MULTIPLE_HEAPS
+        dprintf (3, ("Starting all gc thread after bridge processing"));
+        gc_t_join.restart();
+    }
+#endif //MULTIPLE_HEAPS
+
+    {
+        int thread = heap_number;
+        // Each thread will receive an equal chunk of bridge objects, with the last thread
+        // handling a few more objects from the remainder.
+        size_t count_per_heap = num_global_bridge_objs / n_heaps;
+        size_t start_index = thread * count_per_heap;
+        size_t end_index = (thread == n_heaps - 1) ? num_global_bridge_objs : (thread + 1) * count_per_heap;
+
+        for (size_t obj_idx = start_index; obj_idx < end_index; obj_idx++)
+        {
+            background_mark_simple (global_bridge_list[obj_idx] THREAD_NUMBER_ARG);
+        }
+
+        drain_mark_queue();
+    }
+#endif //FEATURE_JAVAMARSHAL
+
 #ifdef MULTIPLE_HEAPS
     bgc_t_join.join(this, gc_join_null_dead_short_weak);
     if (bgc_t_join.joined())
@@ -49709,6 +49778,11 @@ HRESULT GCHeap::Initialize()
 ////
 // GC callback functions
 bool GCHeap::IsPromoted(Object* object)
+{
+    return IsPromoted(object, true);
+}
+
+bool GCHeap::IsPromoted(Object* object, bool bVerifyNextHeader)
 {
     uint8_t* o = (uint8_t*)object;
 
@@ -49746,10 +49820,11 @@ bool GCHeap::IsPromoted(Object* object)
 #endif //USE_REGIONS
     }
 
+// Walking refs when objects are marked seems unexpected
 #ifdef _DEBUG
     if (o)
     {
-        ((CObjectHeader*)o)->Validate(TRUE, TRUE, is_marked);
+        ((CObjectHeader*)o)->Validate(TRUE, bVerifyNextHeader, is_marked);
 
         // Frozen objects aren't expected to be "not promoted" here
         assert(is_marked || !IsInFrozenSegment(object));
@@ -53227,6 +53302,15 @@ void GCHeap::DiagGetGCSettings(EtwGCSettingsInfo* etw_settings)
 #endif //FEATURE_EVENT_TRACE
 }
 
+void GCHeap::NullBridgeObjectsWeakRefs(size_t length, void* unreachableObjectHandles)
+{
+#ifdef FEATURE_JAVAMARSHAL
+    Ref_NullBridgeObjectsWeakRefs(length, unreachableObjectHandles);
+#else
+    assert(false);
+#endif
+}
+
 #if defined(WRITE_BARRIER_CHECK) && !defined (SERVER_GC)
 // This code is designed to catch the failure to update the write barrier
 // The way it works is to copy the whole heap right after every GC.  The write

src/coreclr/gc/gc.h

@@ -394,4 +394,6 @@ FILE* CreateLogFile(const GCConfigStringHolder& temp_logfile_name, bool is_confi
 
 void log_init_error_to_host (const char* format, ...);
 
+uint64_t GetHighPrecisionTimeStamp();
+
 #endif // __GC_H