Gc enter/leave/poll without HELPER_METHOD_FRAME (#109378)

Rewrite gc polling logic as a managed helper and remove the stress helper. Also move the pinvoke suspension and rare disable paths to take advantage of the QCall infrastructure.

Author: davidwrighton

src/coreclr/System.Private.CoreLib/src/System/Threading/Thread.CoreCLR.cs

@@ -490,5 +490,37 @@ private void ResetFinalizerThreadSlow()
                 Priority = ThreadPriority.Highest;
             }
         }
+
+        [LibraryImport(RuntimeHelpers.QCall, EntryPoint = "ThreadNative_PollGC")]
+        private static partial void ThreadNative_PollGC();
+
+        // GC Suspension is done by simply dropping into native code via p/invoke, and we reuse the p/invoke
+        // mechanism for suspension. On all architectures we should have the actual stub used for the check be implemented
+        // as a small assembly stub which checks the global g_TrapReturningThreads flag and tail-call to this helper
+        private static unsafe void PollGC()
+        {
+            NativeThreadState catchAtSafePoint = ((NativeThreadClass*)Thread.DirectOnThreadLocalData.pNativeThread)->m_State & NativeThreadState.TS_CatchAtSafePoint;
+            if (catchAtSafePoint != NativeThreadState.None)
+            {
+                ThreadNative_PollGC();
+            }
+        }
+
+        [StructLayout(LayoutKind.Sequential)]
+        private struct NativeThreadClass
+        {
+            public NativeThreadState m_State;
+        }
+
+        private enum NativeThreadState
+        {
+            None                      = 0,
+            TS_AbortRequested         = 0x00000001,    // Abort the thread
+            TS_DebugSuspendPending    = 0x00000008,    // Is the debugger suspending threads?
+            TS_GCOnTransitions        = 0x00000010,    // Force a GC on stub transitions (GCStress only)
+
+        // We require (and assert) that the following bits are less than 0x100.
+            TS_CatchAtSafePoint = (TS_AbortRequested | TS_DebugSuspendPending | TS_GCOnTransitions),
+        };
     }
 }

src/coreclr/inc/jithelpers.h

@@ -149,8 +149,8 @@
 
     // GC support
     DYNAMICJITHELPER(CORINFO_HELP_STOP_FOR_GC,  JIT_RareDisableHelper,  METHOD__NIL)
-    JITHELPER(CORINFO_HELP_POLL_GC,             JIT_PollGC,         METHOD__NIL)
-
+    DYNAMICJITHELPER(CORINFO_HELP_POLL_GC,      JIT_PollGC,         METHOD__THREAD__POLLGC)
+    
     JITHELPER(CORINFO_HELP_CHECK_OBJ,           JIT_CheckObj,       METHOD__NIL)
 
     // GC Write barrier support

src/coreclr/vm/amd64/AsmHelpers.asm

@@ -12,6 +12,8 @@ extern  ProfileTailcall:proc
 extern OnHijackWorker:proc
 extern JIT_RareDisableHelperWorker:proc
 
+extern g_pPollGC:QWORD
+extern g_TrapReturningThreads:DWORD
 
 ; EXTERN_C int __fastcall HelperMethodFrameRestoreState(
 ;         INDEBUG_COMMA(HelperMethodFrame *pFrame)
@@ -447,5 +449,13 @@ NESTED_END OnCallCountThresholdReachedStub, _TEXT
 
 endif ; FEATURE_TIERED_COMPILATION
 
-        end
+LEAF_ENTRY JIT_PollGC, _TEXT
+    cmp [g_TrapReturningThreads], 0
+    jnz             JIT_PollGCRarePath
+    ret
+JIT_PollGCRarePath:
+    mov rax, g_pPollGC
+    TAILJMP_RAX
+LEAF_END JIT_PollGC, _TEXT
 
+        end

src/coreclr/vm/amd64/asmhelpers.S

@@ -319,3 +319,14 @@ LEAF_ENTRY GetTlsIndexObjectDescOffset, _TEXT
         int 3
 LEAF_END GetTlsIndexObjectDescOffset, _TEXT
 #endif
+
+LEAF_ENTRY  JIT_PollGC, _TEXT
+        PREPARE_EXTERNAL_VAR g_TrapReturningThreads, rax
+        cmp     dword ptr [rax], 0
+        jnz     LOCAL_LABEL(JIT_PollGCRarePath)
+        ret
+LOCAL_LABEL(JIT_PollGCRarePath):
+        PREPARE_EXTERNAL_VAR g_pPollGC, rax
+        mov     rax, [rax]
+        jmp     rax
+LEAF_END  JIT_PollGC, _TEXT

src/coreclr/vm/appdomain.cpp

@@ -1006,6 +1006,8 @@ extern "C" PCODE g_pGetGCStaticBase;
 PCODE g_pGetGCStaticBase;
 extern "C" PCODE g_pGetNonGCStaticBase;
 PCODE g_pGetNonGCStaticBase;
+extern "C" PCODE g_pPollGC;
+PCODE g_pPollGC;
 
 void SystemDomain::LoadBaseSystemClasses()
 {
@@ -1144,6 +1146,7 @@ void SystemDomain::LoadBaseSystemClasses()
 
         g_pGetGCStaticBase = CoreLibBinder::GetMethod(METHOD__STATICSHELPERS__GET_GC_STATIC)->GetMultiCallableAddrOfCode();
         g_pGetNonGCStaticBase = CoreLibBinder::GetMethod(METHOD__STATICSHELPERS__GET_NONGC_STATIC)->GetMultiCallableAddrOfCode();
+        g_pPollGC = CoreLibBinder::GetMethod(METHOD__THREAD__POLLGC)->GetMultiCallableAddrOfCode();
 
     #ifdef PROFILING_SUPPORTED
         // Note that g_profControlBlock.fBaseSystemClassesLoaded must be set to TRUE only after

src/coreclr/vm/arm/asmhelpers.S

@@ -914,3 +914,26 @@ ProbeLoop:
     NESTED_END OnCallCountThresholdReachedStub, _TEXT
 
 #endif // FEATURE_TIERED_COMPILATION
+
+    LEAF_ENTRY JIT_PollGC, _TEXT
+#if defined(__clang__)
+        ldr     r2, =g_TrapReturningThreads-(1f+4)
+1:
+        add     r2, pc
+#else
+        ldr     r2, =g_TrapReturningThreads
+#endif
+        ldr     r2, [r2]
+        cbnz    r2, LOCAL_LABEL(JIT_PollGCRarePath)
+        bx lr
+LOCAL_LABEL(JIT_PollGCRarePath):
+#if defined(__clang__)
+        ldr     r2, =g_pPollGC-(1f+4)
+1:
+        add     r2, pc
+#else
+        ldr     r2, =g_pPollGC
+#endif
+        ldr     r2, [r2]
+        EPILOG_BRANCH_REG r2
+    LEAF_END JIT_PollGC, _TEXT

src/coreclr/vm/arm64/asmhelpers.S

@@ -806,3 +806,14 @@ LEAF_ENTRY GetTLSResolverAddress, _TEXT
 LEAF_END GetTLSResolverAddress, _TEXT
 // ------------------------------------------------------------------
 #endif // !TARGET_OSX
+
+LEAF_ENTRY  JIT_PollGC, _TEXT
+        PREPARE_EXTERNAL_VAR g_TrapReturningThreads, x9
+        ldr     w9, [x9]
+        cbnz    w9, LOCAL_LABEL(JIT_PollGCRarePath)
+        ret
+LOCAL_LABEL(JIT_PollGCRarePath):
+        PREPARE_EXTERNAL_VAR g_pPollGC, x9
+        ldr     x9, [x9]
+        br x9
+LEAF_END  JIT_PollGC, _TEXT

src/coreclr/vm/arm64/asmhelpers.asm

@@ -39,6 +39,9 @@
     IMPORT  g_pGetGCStaticBase
     IMPORT  g_pGetNonGCStaticBase
 
+    IMPORT g_pPollGC
+    IMPORT g_TrapReturningThreads
+
 #ifdef WRITE_BARRIER_CHECK
     SETALIAS g_GCShadow, ?g_GCShadow@@3PEAEEA
     SETALIAS g_GCShadowEnd, ?g_GCShadowEnd@@3PEAEEA
@@ -1179,6 +1182,17 @@ __HelperNakedFuncName SETS "$helper":CC:"Naked"
 
 #endif ; FEATURE_SPECIAL_USER_MODE_APC
 
+    LEAF_ENTRY  JIT_PollGC
+        ldr     x9, =g_TrapReturningThreads
+        ldr     w9, [x9]
+        cbnz    w9, JIT_PollGCRarePath
+        ret
+JIT_PollGCRarePath
+        ldr     x9, =g_pPollGC
+        ldr     x9, [x9]
+        br x9
+    LEAF_END
+
 
 ; Must be at very end of file
     END

src/coreclr/vm/comsynchronizable.cpp

@@ -848,6 +848,12 @@ extern "C" void QCALLTYPE ThreadNative_DisableComObjectEagerCleanup(QCall::Threa
 }
 #endif //FEATURE_COMINTEROP
 
+extern "C" void QCALLTYPE ThreadNative_PollGC()
+{
+    // This is an intentional no-op.  The call is made to ensure that the thread goes through a GC transition
+    // and is thus marked as a GC safe point, and that the p/invoke rare path will kick in
+}
+
 extern "C" BOOL QCALLTYPE ThreadNative_YieldThread()
 {
     QCALL_CONTRACT;

src/coreclr/vm/comsynchronizable.h

@@ -50,6 +50,7 @@ extern "C" BOOL QCALLTYPE ThreadNative_GetIsBackground(QCall::ThreadHandle threa
 extern "C" void QCALLTYPE ThreadNative_SetIsBackground(QCall::ThreadHandle thread, BOOL value);
 extern "C" void QCALLTYPE ThreadNative_InformThreadNameChange(QCall::ThreadHandle thread, LPCWSTR name, INT32 len);
 extern "C" BOOL QCALLTYPE ThreadNative_YieldThread();
+extern "C" void QCALLTYPE ThreadNative_PollGC();
 extern "C" UINT64 QCALLTYPE ThreadNative_GetCurrentOSThreadId();
 extern "C" void QCALLTYPE ThreadNative_Initialize(QCall::ObjectHandleOnStack t);
 extern "C" INT32 QCALLTYPE ThreadNative_GetThreadState(QCall::ThreadHandle thread);