// Licensed to the .NET Foundation under one or more agreements. // The .NET Foundation licenses this file to you under the MIT license. // // // Currently represents a logical and physical CLR thread. Later, these concepts will be separated. // // // #SuspendingTheRuntime // // One of the primary differences between runtime code (managed code), and traditional (unmanaged code) is // the existence of the GC heap (see file:gc.cpp#Overview). For the GC to do its job, it must be able to // traverse all references to the GC heap, including ones on the stack of every thread, as well as any in // hardware registers. While it is simple to state this requirement, it has long reaching effects, because // properly accounting for all GC heap references ALL the time turns out to be quite hard. When we make a // bookkeeping mistake, a GC reference is not reported at GC time, which means it will not be updated when the // GC happens. Since memory in the GC heap can move, this can cause the pointer to point at 'random' places // in the GC heap, causing data corruption. This is a 'GC Hole', and is very bad. We have special modes (see // code:EEConfig.GetGCStressLevel) called GCStress to help find such issues. // // In order to find all GC references on the stacks, we need to ensure that no thread is manipulating a GC // reference at the time of the scan. This is the job of code:Thread.SuspendRuntime. Logically, it suspends // every thread in the process. Unfortunately, it can not literally simply call the OS SuspendThread API on // all threads. The reason is that the other threads MIGHT hold important locks (for example, there is a lock // that is taken when unmanaged heap memory is requested, or when a DLL is loaded). In general, process // global structures in the OS will be protected by locks, and if you suspend a thread it might hold that // lock. If you happen to need that OS service (eg you might need to allocate unmanaged memory), then // deadlock will occur (as you wait on the suspended thread, that never wakes up). // // Luckily, we don't need to actually suspend the threads, we just need to ensure that all GC references on // the stack are stable. This is where the concept of cooperative mode and preemptive mode (a bad name) come // from. // // #CooperativeMode // // The runtime keeps a table of all threads that have ever run managed code in the code:ThreadStore table. // The ThreadStore table holds a list of Thread objects (see code:#ThreadClass). This object holds all // information about managed threads. Cooperative mode is defined as the mode the thread is in when the field // code:Thread.m_fPreemptiveGCDisabled is non-zero. When this field is zero, the thread is said to be in // Preemptive mode (named because if you preempt the thread in this mode, it is guaranteed to be in a place // where a GC can occur). // // When a thread is in cooperative mode, it is basically saying that it is potentially modifying GC // references, and so the runtime must Cooperate with it to get to a 'GC Safe' location where the GC // references can be enumerated. This is the mode that a thread is in MOST times when it is running managed // code (in fact, if the EIP is in JIT compiled code, there is only one place where you are NOT in cooperative // mode (Inlined PINVOKE transition code)). Conversely, any time non-runtime unmanaged code is running, the // thread MUST NOT be in cooperative mode (you risk deadlock otherwise). Only code in mscorwks.dll might be // running in either cooperative or preemptive mode. // // It is easier to describe the invariant associated with being in Preemptive mode. When the thread is in // preemptive mode (when code:Thread.m_fPreemptiveGCDisabled is zero), the thread guarantees two things // // * That it is not currently running code that manipulates GC references. // * That it has set the code:Thread.m_pFrame pointer in the code:Thread to be a subclass of the class // code:Frame which marks the location on the stack where the last managed method frame is. This // allows the GC to start crawling the stack from there (essentially skip over the unmanaged frames). // * That the thread will not reenter managed code if the global variable code:g_TrapReturningThreads is // set (it will call code:Thread.RareDisablePreemptiveGC first which will block if a suspension is // in progress) // // The basic idea is that the suspension logic in code:Thread.SuspendRuntime first sets the global variable // code:g_TrapReturningThreads and then checks if each thread in the ThreadStore is in Cooperative mode. If a // thread is NOT in cooperative mode, the logic simply skips the thread, because it knows that the thread // will stop itself before reentering managed code (because code:g_TrapReturningThreads is set). This avoids // the deadlock problem mentioned earlier, because threads that are running unmanaged code are allowed to // run. Enumeration of GC references starts at the first managed frame (pointed at by code:Thread.m_pFrame). // // When a thread is in cooperative mode, it means that GC references might be in the process of being // manipulated. There are two important possibilities // // * The CPU is running JIT compiled code // * The CPU is running code elsewhere (which should only be in mscorwks.dll, because everywhere else a // transition to preemptive mode should have happened first) // // * #PartiallyInterruptibleCode // * #FullyInterruptibleCode // // If the Instruction pointer (x86/x64: EIP, ARM: R15/PC) is in JIT compiled code, we can detect this because we have tables that // map the ranges of every method back to their code:MethodDesc (this the code:ICodeManager interface). In // addition to knowing the method, these tables also point at 'GCInfo' that tell for that method which stack // locations and which registers hold GC references at any particular instruction pointer. If the method is // what is called FullyInterruptible, then we have information for any possible instruction pointer in the // method and we can simply stop the thread (however we have to do this carefully TODO explain). // // However for most methods, we only keep GC information for particular EIPs, in particular we keep track of // GC reference liveness only at call sites. Thus, not every location is 'GC Safe' (that is, we can enumerate // all references, but must be 'driven' to a GC safe location). // // We drive threads to GC safe locations by hijacking. This is a term for updating the return address on the // stack so that we gain control when a method returns. If we find that we are in JITTed code but NOT at a GC // safe location, then we find the return address for the method and modify it to cause the runtime to stop. // We then let the method run. Hopefully the method quickly returns, and hits our hijack, and we are now at a // GC-safe location (all call sites are GC-safe). If not, we repeat the procedure (possibly moving the // hijack). At some point, a method returns, and we get control. For methods that have loops that don't make // calls, we are forced to make the method FullyInterruptible, so we can be sure to stop the method. // // This leaves only the case where we are in cooperative modes, but not in JIT compiled code (we should be in // clr.dll). In this case, we simply let the thread run. The idea is that code in clr.dll makes the // promise that it will not do ANYTHING that will block (which includes taking a lock), while in cooperative // mode, or do anything that might take a long time without polling to see if a GC is needed. Thus, this code // 'cooperates' to ensure that GCs can happen in a timely fashion. // // If you need to switch the GC mode of the current thread, look for the GCX_COOP() and GCX_PREEMP() macros. // #ifndef __threads_h__ #define __threads_h__ #include #include "vars.hpp" #include "util.hpp" #include "argslot.h" #include "regdisp.h" #include "mscoree.h" #include "gcheaputilities.h" #include "gchandleutilities.h" #include "gcinfotypes.h" #include #include "cdacdata.h" class Thread; class ThreadStore; class MethodDesc; struct PendingSync; class AppDomain; class PInvoke; class Frame; class ThreadBaseObject; class AppDomainStack; class DeadlockAwareLock; class EECodeInfo; class DebuggerPatchSkip; class FaultingExceptionFrame; enum BinderMethodID : int; class PrepareCodeConfig; class NativeCodeVersion; struct InterpThreadContext; typedef void(*ADCallBackFcnType)(LPVOID); #include "stackwalktypes.h" #include "log.h" #include "excep.h" #include "synch.h" #include "exstate.h" #include "interoputil.h" #include "eventtrace.h" #ifdef FEATURE_PERFTRACING #include "eventpipeadaptertypes.h" #endif // FEATURE_PERFTRACING #include "threadstatics.h" class Module; // TailCallArgBuffer states #define TAILCALLARGBUFFER_ACTIVE 0 #define TAILCALLARGBUFFER_INSTARG_ONLY 1 #define TAILCALLARGBUFFER_INACTIVE 2 struct TailCallArgBuffer { int State; int Size; void* GCDesc; BYTE Args[1]; }; #if (defined(TARGET_ARM) && defined(FEATURE_EMULATE_SINGLESTEP)) #include "armsinglestepper.h" #endif #if (defined(TARGET_ARM64) && defined(FEATURE_EMULATE_SINGLESTEP)) #include "arm64singlestepper.h" #endif #if (defined(TARGET_RISCV64) && defined(FEATURE_EMULATE_SINGLESTEP)) #include "riscv64singlestepper.h" #endif #if (defined(TARGET_LOONGARCH64) && defined(FEATURE_EMULATE_SINGLESTEP)) #include "loongarch64singlestepper.h" #endif #if !defined(PLATFORM_SUPPORTS_SAFE_THREADSUSPEND) // DISABLE_THREADSUSPEND controls whether Thread::SuspendThread will be used at all. // This API is dangerous on non-Windows platforms, as it can lead to deadlocks, // due to low level OS resources that the PAL is not aware of, or due to the fact that // PAL-unaware code in the process may hold onto some OS resources. #define DISABLE_THREADSUSPEND #endif #if defined(FEATURE_HIJACK) && (defined(TARGET_UNIX) || defined(FEATURE_SPECIAL_USER_MODE_APC)) #define FEATURE_THREAD_ACTIVATION #endif // NT thread priorities range from -15 to +15. #define INVALID_THREAD_PRIORITY ((DWORD)0x80000000) // For a fiber which switched out, we set its OSID to a special number // Note: there's a copy of this macro in strike.cpp #define SWITCHED_OUT_FIBER_OSID 0xbaadf00d; #ifdef _DEBUG // A thread doesn't receive its id until fully constructed. #define UNINITIALIZED_THREADID 0xbaadf00d #endif //_DEBUG //*************************************************************************** #ifdef FEATURE_HIJACK // Used to capture information about the state of execution of a *SUSPENDED* thread. struct ExecutionState; #ifndef TARGET_UNIX // This is the type of the start function of a redirected thread pulled from // a HandledJITCase during runtime suspension typedef void (__stdcall *PFN_REDIRECTTARGET)(); // Describes the weird argument sets during hijacking struct HijackArgs; #endif // !TARGET_UNIX #endif // FEATURE_HIJACK // manifest constant for waiting in the exposed classlibs const INT32 INFINITE_TIMEOUT = -1; /***************************************************************************/ #ifdef FEATURE_SPECIAL_USER_MODE_APC // These declarations are for a new special user-mode APC feature introduced in Windows. These are not yet available in Windows // SDK headers, so some names below are prefixed with "CLONE_" to avoid conflicts in the future. Once the prefixed declarations // become available in the Windows SDK headers, the prefixed declarations below can be removed in favor of the SDK ones. enum CLONE_QUEUE_USER_APC_FLAGS { CLONE_QUEUE_USER_APC_FLAGS_NONE = 0x0, CLONE_QUEUE_USER_APC_FLAGS_SPECIAL_USER_APC = 0x1, CLONE_QUEUE_USER_APC_CALLBACK_DATA_CONTEXT = 0x10000 }; struct CLONE_APC_CALLBACK_DATA { ULONG_PTR Parameter; PCONTEXT ContextRecord; ULONG_PTR Reserved0; ULONG_PTR Reserved1; }; typedef CLONE_APC_CALLBACK_DATA *CLONE_PAPC_CALLBACK_DATA; typedef BOOL (WINAPI *QueueUserAPC2Proc)(PAPCFUNC ApcRoutine, HANDLE Thread, ULONG_PTR Data, CLONE_QUEUE_USER_APC_FLAGS Flags); const CLONE_QUEUE_USER_APC_FLAGS SpecialUserModeApcWithContextFlags = (CLONE_QUEUE_USER_APC_FLAGS) ( CLONE_QUEUE_USER_APC_FLAGS_SPECIAL_USER_APC | // this will be a special user-mode APC CLONE_QUEUE_USER_APC_CALLBACK_DATA_CONTEXT // the callback's parameter will be a PAPC_CALLBACK_DATA ); #endif // FEATURE_SPECIAL_USER_MODE_APC //*************************************************************************** // Public functions // // Thread* GetThread() - returns current Thread. // Thread* SetupThread() - creates a new Thread. // Thread* SetupThreadNoThrow() - creates a new Thread without throwing. // Thread* SetupUnstartedThread() - creates new unstarted Thread which // (obviously) isn't in a TLS. // void DestroyThread() - the underlying logical thread is going // away. // void DetachThread() - the underlying logical thread is going // away but we don't want to destroy it yet. // // Public functions for one-time init/cleanup // // void InitThreadManager() - onetime init // void TerminateThreadManager() - onetime cleanup // // Public functions for taking control of a thread at a safe point // // VOID OnHijackTripThread() - we've hijacked a JIT method // VOID OnHijackFPTripThread() - we've hijacked a JIT method, // and need to save the x87 FP stack. // //*************************************************************************** //*************************************************************************** // Public functions //*************************************************************************** //--------------------------------------------------------------------------- // //--------------------------------------------------------------------------- Thread* SetupThread(); Thread* SetupThreadNoThrow(HRESULT *phresult = NULL); enum SetupUnstartedThreadFlags { SUTF_None = 0, // The ThreadStoreLock is being held during Thread startup. SUTF_ThreadStoreLockAlreadyTaken = 1, // The default flags for the majority of threads. SUTF_Default = SUTF_None, }; Thread* SetupUnstartedThread(SetupUnstartedThreadFlags flags = SUTF_Default); void DestroyThread(Thread *th); DWORD GetRuntimeId(); #define CREATETHREAD_IF_NULL_FAILFAST(thread__, msg__) \ { \ HRESULT ctinffhr__; \ thread__ = SetupThreadNoThrow(&ctinffhr__); \ if (thread__ == NULL) \ { \ EEPOLICY_HANDLE_FATAL_ERROR_WITH_MESSAGE(ctinffhr__, msg__); \ UNREACHABLE(); \ } \ } //--------------------------------------------------------------------------- // One-time initialization. Called during Dll initialization. //--------------------------------------------------------------------------- void InitThreadManager(); void InitThreadManagerTracingData(); #ifndef FEATURE_PORTABLE_HELPERS void ReportCopiedWriteBarriersToPerfMap(); #ifdef FEATURE_EVENT_TRACE void ReportCopiedWriteBarriersToEventTracing(DWORD eventOptions); #endif // FEATURE_EVENT_TRACE #endif // !FEATURE_PORTABLE_HELPERS // When we want to take control of a thread at a safe point, the thread will // eventually come back to us in one of the following trip functions: #ifdef FEATURE_HIJACK EXTERN_C void WINAPI OnHijackTripThread(); #ifdef TARGET_X86 EXTERN_C void WINAPI OnHijackFPTripThread(); // hijacked JIT code is returning an FP value #endif // TARGET_X86 #endif // FEATURE_HIJACK void CommonTripThread(); void SetupTLSForThread(); // When we resume a thread at a new location, to get an exception thrown, we have to // pretend the exception originated elsewhere. EXTERN_C void ThrowControlForThread( #if !defined(TARGET_X86) FaultingExceptionFrame *pfef #endif // !TARGET_X86 #if defined(TARGET_AMD64) && defined(TARGET_WINDOWS) , TADDR ssp #endif // TARGET_AMD64 && TARGET_WINDOWS ); #if defined(_DEBUG) BOOL MatchThreadHandleToOsId ( HANDLE h, DWORD osId ); #endif // The Thread class represents a managed thread. This thread could be internal // or external (i.e. it wandered in from outside the runtime). For internal // threads, it could correspond to an exposed System.Thread object or it // could correspond to an internal worker thread of the runtime. // // If there's a physical Win32 thread underneath this object (i.e. it isn't an // unstarted System.Thread), then this instance can be found in the TLS // of that physical thread. #ifdef FEATURE_HIJACK // Hijack function returning EXTERN_C void STDCALL OnHijackWorker(HijackArgs * pArgs); #endif // FEATURE_HIJACK // This is the code we pass around for Thread.Interrupt, mainly for assertions #define APC_Code 0xEECEECEE #ifdef DACCESS_COMPILE class BaseStackGuard; #endif struct PortableTailCallFrame { void* TailCallAwareReturnAddress; void* NextCall; }; class TailCallTls { friend class CoreLibBinder; PortableTailCallFrame* m_frame; TailCallArgBuffer* m_argBuffer; public: TailCallTls(); TailCallArgBuffer* AllocArgBuffer(int size); void FreeArgBuffer() { delete[] (BYTE*)m_argBuffer; m_argBuffer = NULL; } TailCallArgBuffer* GetArgBuffer() { return m_argBuffer; } const PortableTailCallFrame* GetFrame() { return m_frame; } }; // This struct contains data that lives as long as the current OS thread. struct RuntimeThreadLocals { // on MP systems, each thread has its own allocation chunk so we can avoid // lock prefixes and expensive MP cache snooping stuff ee_alloc_context alloc_context; }; #ifdef _MSC_VER // use selectany to avoid initialization de-optimization issues in the compiler __declspec(selectany) #else extern #endif PLATFORM_THREAD_LOCAL RuntimeThreadLocals t_runtime_thread_locals; typedef DPTR(struct RuntimeThreadLocals) PTR_RuntimeThreadLocals; typedef DPTR(struct gc_alloc_context) PTR_gc_alloc_context; // #ThreadClass // // A code:Thread contains all the per-thread information needed by the runtime. We can get this // structure through the OS TLS slot see code:#RuntimeThreadLocals for more information. class Thread { friend class ThreadStore; friend class ThreadSuspend; friend class SyncBlock; friend struct PendingSync; #ifdef _DEBUG friend class EEContract; #endif #ifdef DACCESS_COMPILE friend class ClrDataAccess; friend class ClrDataTask; #endif friend BOOL NTGetThreadContext(Thread *pThread, T_CONTEXT *pContext); friend BOOL NTSetThreadContext(Thread *pThread, const T_CONTEXT *pContext); #ifdef FEATURE_HIJACK // MapWin32FaultToCOMPlusException needs access to Thread::IsAddrOfRedirectFunc() friend DWORD MapWin32FaultToCOMPlusException(EXCEPTION_RECORD *pExceptionRecord); friend void STDCALL OnHijackWorker(HijackArgs * pArgs); #ifdef FEATURE_THREAD_ACTIVATION friend void HandleSuspensionForInterruptedThread(CONTEXT *interruptedContext); friend BOOL CheckActivationSafePoint(SIZE_T ip); #endif // FEATURE_THREAD_ACTIVATION #endif // FEATURE_HIJACK friend void InitThreadManager(); // Debug and Profiler caches ThreadHandle. friend class Debugger; // void Debugger::ThreadStarted(Thread* pRuntimeThread, BOOL fAttaching); #if defined(DACCESS_COMPILE) friend class DacDbiInterfaceImpl; // DacDbiInterfaceImpl::GetThreadHandle(HANDLE * phThread); #endif // DACCESS_COMPILE friend class ProfToEEInterfaceImpl; // HRESULT ProfToEEInterfaceImpl::GetHandleFromThread(ThreadID threadId, HANDLE *phThread); friend class ThreadExceptionState; friend class StackFrameIterator; public: enum SetThreadStackGuaranteeScope { STSGuarantee_Force, STSGuarantee_OnlyIfEnabled }; static BOOL IsSetThreadStackGuaranteeInUse(SetThreadStackGuaranteeScope fScope = STSGuarantee_OnlyIfEnabled) { WRAPPER_NO_CONTRACT; if(STSGuarantee_Force == fScope) return TRUE; #ifdef DEBUG // For debug, always enable setting thread stack guarantee so that we can print the stack trace return TRUE; #else return FALSE; #endif } public: // If we are trying to suspend a thread, we set the appropriate pending bit to // indicate why we want to suspend it (TS_AbortRequested or TS_DebugSuspendPending). // // If instead the thread has blocked itself, via WaitForDebugSuspend, we indicate // this with TS_DebugSyncSuspended. A user request is not allowed to resume a thread // suspended for debugging. -- That's not strictly true. It is allowed to resume such a // thread so long as it was ALSO suspended by the user. In other words, this // ensures that user resumptions aren't unbalanced from user suspensions. // enum ThreadState { TS_Unknown = 0x00000000, // threads are initialized this way. [cDAC] [Thread]: Contract depends on this value. TS_AbortRequested = 0x00000001, // Abort the thread TS_SuspensionTrapped = 0x00000002, // Thread is trapped waiting for suspension to complete (was in managed code). [cDAC] [Thread]: Contract depends on this value. TS_GCSuspendRedirected = 0x00000004, // Thread has been redirected to suspension routine. [cDAC] [Thread]: Contract depends on this value. TS_DebugSuspendPending = 0x00000008, // Is the debugger suspending threads? [cDAC] [Thread]: Contract depends on this value. TS_GCOnTransitions = 0x00000010, // Force a GC on stub transitions (GCStress only) TS_SyncBlockCleanup = 0x00000020, // The synch block needs to be cleaned up. TS_ExecutingOnAltStack = 0x00000040, // Runtime is executing on an alternate stack located anywhere in the memory #ifdef FEATURE_HIJACK TS_Hijacked = 0x00000080, // Return address has been hijacked. [cDAC] [Thread]: Contract depends on this value. #endif // FEATURE_HIJACK // unused = 0x00000100, TS_Background = 0x00000200, // Thread is a background thread. [cDAC] [Thread]: Contract depends on this value. TS_Unstarted = 0x00000400, // Thread has never been started. [cDAC] [Thread]: Contract depends on this value. TS_Dead = 0x00000800, // .NET runtime has finished shutting down this thread, and it is about to be terminated by the OS. TS_WeOwn = 0x00001000, // Exposed object initiated this thread #ifdef FEATURE_COMINTEROP_APARTMENT_SUPPORT TS_CoInitialized = 0x00002000, // CoInitialize has been called for this thread. [cDAC] [Thread]: Contract depends on this value. TS_InSTA = 0x00004000, // Thread hosts an STA. [cDAC] [Thread]: Contract depends on this value. TS_InMTA = 0x00008000, // Thread is part of the MTA. [cDAC] [Thread]: Contract depends on this value. #endif // FEATURE_COMINTEROP_APARTMENT_SUPPORT // Some bits that only have meaning for reporting the state to clients. TS_Stopped = 0x00010000, // Thread has started to shut down and should not run managed code. Equivalent to ThreadState.Stopped. [cDAC] [Thread]: Contract depends on this value. TS_FullyInitialized = 0x00020000, // Thread is fully initialized and we are ready to broadcast its existence to external clients // unused = 0x00040000, TS_DebugSyncSuspended = 0x00080000, // Thread has suspended itself at a safe point in response to a debugger suspend request. [cDAC] [Thread]: Contract depends on this value. TS_DebugWillSync = 0x00100000, // Debugger will wait for this thread to sync. [cDAC] [Thread]: Contract depends on this value. TS_StackCrawlNeeded = 0x00200000, // A stackcrawl is needed on this thread, such as for thread abort // See comment for s_pWaitForStackCrawlEvent for reason. // unused = 0x00400000, // unused = 0x00800000, TS_TPWorkerThread = 0x01000000, // is this a threadpool worker thread? [cDAC] [Thread]: Contract depends on this value. TS_WaitSleepJoin = 0x02000000, // sitting in a Sleep(), Wait(), Join(). [cDAC] [Thread]: Contract depends on this value. TS_Interrupted = 0x04000000, // was awakened by an interrupt APC. !!! This can be moved to TSNC // unused TS_AbortInitiated = 0x10000000, // set when abort is begun TS_Finalized = 0x20000000, // The associated managed Thread object has been finalized. // We can clean up the unmanaged part now. TS_FailStarted = 0x40000000, // The thread fails during startup. TS_Detached = 0x80000000, // Thread was detached by DllMain. [cDAC] [Thread]: Contract depends on this value. // @TODO: We need to reclaim the bits that have no concurrency issues (i.e. they are only // manipulated by the owning thread) and move them off to a different DWORD. Note if this // enum is changed, we also need to update SOS to reflect this. // We require (and assert) that the following bits are less than 0x100. TS_CatchAtSafePoint = (TS_AbortRequested | TS_DebugSuspendPending | TS_GCOnTransitions), }; // Thread flags that have no concurrency issues (i.e., they are only manipulated by the owning thread). Use these // state flags when you have a new thread state that doesn't belong in the ThreadState enum above. enum ThreadStateNoConcurrency { TSNC_Unknown = 0x00000000, // threads are initialized this way TSNC_DebuggerThreadStartSent = 0x00000001, // The debugger thread-start event has been sent for this thread. // unused = 0x00000002, TSNC_DebuggerIsStepping = 0x00000004, // debugger is stepping this thread TSNC_DebuggerIsManagedException = 0x00000008, // EH is re-raising a managed exception. // unused = 0x00000010, TSNC_BlockedForShutdown = 0x00000020, // Thread is blocked in WaitForEndOfShutdown. We should not hit WaitForEndOfShutdown again. // unused = 0x00000040, // unused = 0x00000080, TSNC_ExistInThreadStore = 0x00000100, // For dtor to know if it needs to be removed from ThreadStore // unused = 0x00000200, TSNC_OwnsSpinLock = 0x00000400, // The thread owns a spinlock. TSNC_PreparingAbort = 0x00000800, // Preparing abort. This avoids recursive HandleThreadAbort call. // unused = 0x00001000, // unused = 0x00002000, TSNC_CreatingTypeInitException = 0x00004000, // Thread is trying to create a TypeInitException // unused = 0x00008000, // unused = 0x00010000, TSNC_InRestoringSyncBlock = 0x00020000, // The thread is restoring its SyncBlock for Object.Wait. // After the thread is interrupted once, we turn off interruption // at the beginning of wait. // unused = 0x00040000, // unused = 0x00080000, // unused = 0x00100000, // unused = 0x00200000, // unused = 0x00400000, // unused = 0x00800000, TSNC_ProcessedUnhandledException = 0x01000000,// Set on a thread on which we have done unhandled exception processing so that // we dont perform it again when OS invokes our UEF. Currently, applicable threads include: // 1) entry point thread of a managed app // 2) new managed thread created in default domain // // For such threads, we will return to the OS after our UE processing is done // and the OS will start invoking the UEFs. If our UEF gets invoked, it will try to // perform the UE processing again. We will use this flag to prevent the duplicated // effort. // // Once we are completely independent of the OS UEF, we could remove this. TSNC_SkipManagedPersonalityRoutine = 0x02000000, // Ignore the ProcessCLRException calls when propagating exception to external native code // unused = 0x04000000, // unused = 0x08000000, TSNC_TSLTakenForStartup = 0x10000000, // The ThreadStoreLock (TSL) is held by another mechanism during // thread startup so can be skipped. TSNC_CallingManagedCodeDisabled = 0x20000000, // Use by multicore JIT feature to asert on calling managed code/loading module in background thread // Exception, system module is allowed, security demand is allowed TSNC_LoadsTypeViolation = 0x40000000, // Use by type loader to break deadlocks caused by type load level ordering violations TSNC_EtwStackWalkInProgress = 0x80000000, // Set on the thread so that ETW can know that stackwalking is in progress // and does not proceed with a stackwalk on the same thread // There are cases during managed debugging when we can run into this situation }; // Thread state flags that are only written by the debugger (out-of-proc) and read by the runtime (in-proc). // Separated from ThreadStateNoConcurrency to avoid read-modify-write races between the debugger and the runtime. enum DebuggerControlledThreadState { DCTS_None = 0x00000000, // [cDAC] [Thread]: Contract depends on this value. DCTS_UserSuspend = 0x00000001, // Marked "suspended" by the debugger [cDAC] [Thread]: Contract depends on this value. }; public: HRESULT DetachThread(BOOL inTerminationCallback); void SetThreadState(ThreadState ts) { LIMITED_METHOD_CONTRACT; InterlockedOr((LONG*)&m_State, ts); } void ResetThreadState(ThreadState ts) { LIMITED_METHOD_CONTRACT; InterlockedAnd((LONG*)&m_State, ~ts); } BOOL HasThreadState(ThreadState ts) { LIMITED_METHOD_CONTRACT; return ((DWORD)m_State & ts); } // // This is meant to be used for quick opportunistic checks for thread abort and similar conditions. This method // does not erect memory barrier and so it may return wrong result sometime that the caller has to handle. // BOOL HasThreadStateOpportunistic(ThreadState ts) { LIMITED_METHOD_CONTRACT; return m_State.LoadWithoutBarrier() & ts; } void SetThreadStateNC(ThreadStateNoConcurrency tsnc) { LIMITED_METHOD_CONTRACT; m_StateNC = (ThreadStateNoConcurrency)((DWORD)m_StateNC | tsnc); } void ResetThreadStateNC(ThreadStateNoConcurrency tsnc) { LIMITED_METHOD_CONTRACT; m_StateNC = (ThreadStateNoConcurrency)((DWORD)m_StateNC & ~tsnc); } BOOL HasThreadStateNC(ThreadStateNoConcurrency tsnc) { LIMITED_METHOD_DAC_CONTRACT; return ((DWORD)m_StateNC & tsnc); } void SetDebuggerControlledThreadState(DebuggerControlledThreadState dcts) { LIMITED_METHOD_CONTRACT; m_DebuggerControlledThreadState = (DebuggerControlledThreadState)((DWORD)m_DebuggerControlledThreadState.Load() | dcts); } void ResetDebuggerControlledThreadState(DebuggerControlledThreadState dcts) { LIMITED_METHOD_CONTRACT; m_DebuggerControlledThreadState = (DebuggerControlledThreadState)((DWORD)m_DebuggerControlledThreadState.Load() & ~dcts); } BOOL HasDebuggerControlledThreadState(DebuggerControlledThreadState dcts) { LIMITED_METHOD_DAC_CONTRACT; return ((DWORD)m_DebuggerControlledThreadState.Load() & dcts); } void MarkEtwStackWalkInProgress() { WRAPPER_NO_CONTRACT; SetThreadStateNC(Thread::TSNC_EtwStackWalkInProgress); } void MarkEtwStackWalkCompleted() { WRAPPER_NO_CONTRACT; ResetThreadStateNC(Thread::TSNC_EtwStackWalkInProgress); } BOOL IsEtwStackWalkInProgress() { WRAPPER_NO_CONTRACT; return HasThreadStateNC(Thread::TSNC_EtwStackWalkInProgress); } DWORD RequireSyncBlockCleanup() { LIMITED_METHOD_CONTRACT; return (m_State & TS_SyncBlockCleanup); } void SetSyncBlockCleanup() { LIMITED_METHOD_CONTRACT; InterlockedOr((LONG*)&m_State, TS_SyncBlockCleanup); } void ResetSyncBlockCleanup() { LIMITED_METHOD_CONTRACT; InterlockedAnd((LONG*)&m_State, ~TS_SyncBlockCleanup); } #ifdef FEATURE_COMINTEROP_APARTMENT_SUPPORT DWORD IsCoInitialized() { LIMITED_METHOD_CONTRACT; return (m_State & TS_CoInitialized); } void SetCoInitialized() { LIMITED_METHOD_CONTRACT; InterlockedOr((LONG*)&m_State, TS_CoInitialized); } void ResetCoInitialized() { LIMITED_METHOD_CONTRACT; ResetThreadState(TS_CoInitialized); } void CleanupCOMState(); #endif // FEATURE_COMINTEROP_APARTMENT_SUPPORT void PrepareApartmentAndContext(); #ifdef FEATURE_COMINTEROP bool IsDisableComObjectEagerCleanup() { LIMITED_METHOD_CONTRACT; return m_fDisableComObjectEagerCleanup; } void SetDisableComObjectEagerCleanup() { LIMITED_METHOD_CONTRACT; m_fDisableComObjectEagerCleanup = true; } #endif //FEATURE_COMINTEROP #ifndef DACCESS_COMPILE bool HasDeadThreadBeenConsideredForGCTrigger() { LIMITED_METHOD_CONTRACT; _ASSERTE(IsDead()); return m_fHasDeadThreadBeenConsideredForGCTrigger; } void SetHasDeadThreadBeenConsideredForGCTrigger() { LIMITED_METHOD_CONTRACT; _ASSERTE(IsDead()); m_fHasDeadThreadBeenConsideredForGCTrigger = true; } #endif // !DACCESS_COMPILE #ifndef DACCESS_COMPILE DWORD CatchAtSafePoint() { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_COOPERATIVE; } CONTRACTL_END; return g_TrapReturningThreads & 1 || HasThreadStateOpportunistic(TS_CatchAtSafePoint); } #endif // DACCESS_COMPILE DWORD IsExecutingOnAltStack() { LIMITED_METHOD_CONTRACT; return (m_State & TS_ExecutingOnAltStack); } void SetExecutingOnAltStack() { LIMITED_METHOD_CONTRACT; SetThreadState(TS_ExecutingOnAltStack); } DWORD IsBackground() { LIMITED_METHOD_CONTRACT; return (m_State & TS_Background); } DWORD IsUnstarted() { LIMITED_METHOD_CONTRACT; SUPPORTS_DAC; return (m_State & TS_Unstarted); } DWORD IsDead() { LIMITED_METHOD_CONTRACT; return (m_State & TS_Dead); } DWORD DoWeOwn() { LIMITED_METHOD_CONTRACT; return (m_State & TS_WeOwn); } ThreadState GetState() { LIMITED_METHOD_CONTRACT; return m_State; } // For delayed destruction of threads DWORD IsDetached() { LIMITED_METHOD_CONTRACT; return (m_State & TS_Detached); } static LONG m_DetachCount; static LONG m_ActiveDetachCount; // Count how many non-background detached // Offsets for the following variables need to fit in 1 byte, so keep near // the top of the object. Also, we want cache line filling to work for us // so the critical stuff is ordered based on frequency of use. Volatile m_State; // Bits for the state of the thread // If TRUE, GC is scheduled cooperatively with this thread. // NOTE: This "byte" is actually a boolean - we don't allow // recursive disables. Volatile m_fPreemptiveGCDisabled; PTR_Frame m_pFrame; // The Current Frame // Unique thread id used for thin locks - kept as small as possible, as we have limited space // in the object header to store it. DWORD m_ThreadId; #ifndef DACCESS_COMPILE Frame* NotifyFrameChainOfExceptionUnwind(Frame* pStartFrame, LPVOID pvLimitSP); #endif // DACCESS_COMPILE // Lock thread is trying to acquire VolatilePtr m_pBlockingLock; // We store a pointer to the runtime thread locals here for easier introspection // from other threads and diagnostic tools PTR_RuntimeThreadLocals m_pRuntimeThreadLocals; public: inline void InitRuntimeThreadLocals() { LIMITED_METHOD_CONTRACT; m_pRuntimeThreadLocals = PTR_RuntimeThreadLocals(&t_runtime_thread_locals); } inline ee_alloc_context* GetEEAllocContext() { LIMITED_METHOD_CONTRACT; if (m_pRuntimeThreadLocals == nullptr) { return nullptr; } return &m_pRuntimeThreadLocals->alloc_context; } inline gc_alloc_context* GetAllocContext() { LIMITED_METHOD_CONTRACT; if (m_pRuntimeThreadLocals == nullptr) { return nullptr; } return &m_pRuntimeThreadLocals->alloc_context.m_GCAllocContext; } // This is the type handle of the first object in the alloc context at the time // we fire the AllocationTick event. It's only for tooling purpose. TypeHandle m_thAllocContextObj; #ifdef FEATURE_INTERPRETER public: PTR_InterpThreadContext m_pInterpThreadContext; InterpThreadContext* GetInterpThreadContext(); InterpThreadContext* GetOrCreateInterpThreadContext(); #endif // FEATURE_INTERPRETER inline void SetTHAllocContextObj(TypeHandle th) {LIMITED_METHOD_CONTRACT; m_thAllocContextObj = th; } inline TypeHandle GetTHAllocContextObj() {LIMITED_METHOD_CONTRACT; return m_thAllocContextObj; } // Flags for thread states that have no concurrency issues. ThreadStateNoConcurrency m_StateNC; // Flags for thread states controlled by the debugger. Volatile m_DebuggerControlledThreadState; private: #ifdef _DEBUG DWORD dbg_m_cSuspendedThreads; // Count of suspended threads that we know are not in native code (and therefore cannot hold OS lock which prevents us calling out to host) DWORD dbg_m_cSuspendedThreadsWithoutOSLock; EEThreadId m_Creator; #endif // A thread may forbid its own suspension. For example when holding certain locks. // The state is a counter to allow nested forbids. // The state is only modified by the current thread. // Other threads may read this state, but must mind the races. // It must be assumed that a running thread (or one that may start running) may change this state at any time. // One exception: SuspendThread can read this "reliably" from other threads by temporarily suspending them, reading, and releasing if != 0. Volatile m_dwForbidSuspendThread; public: static void IncForbidSuspendThread() { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_ANY; SUPPORTS_DAC; } CONTRACTL_END; #ifndef DACCESS_COMPILE Thread * pThread = GetThreadNULLOk(); if (pThread) { _ASSERTE (pThread->m_dwForbidSuspendThread != (LONG)MAXLONG); #ifdef _DEBUG { //DEBUG_ONLY; STRESS_LOG2(LF_SYNC, LL_INFO100000, "Set forbid suspend [%d] for thread %p.\n", pThread->m_dwForbidSuspendThread.Load(), pThread); } #endif // modified only by the current thread, so ++ is ok pThread->m_dwForbidSuspendThread.RawValue()++; } #endif //!DACCESS_COMPILE } static void DecForbidSuspendThread() { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_ANY; SUPPORTS_DAC; } CONTRACTL_END; #ifndef DACCESS_COMPILE Thread * pThread = GetThreadNULLOk(); if (pThread) { _ASSERTE (pThread->m_dwForbidSuspendThread >= (LONG)0); // modified only by the current thread, so -- is ok pThread->m_dwForbidSuspendThread.RawValue()--; #ifdef _DEBUG { //DEBUG_ONLY; STRESS_LOG2(LF_SYNC, LL_INFO100000, "Reset forbid suspend [%d] for thread %p.\n", pThread->m_dwForbidSuspendThread.Load(), pThread); } #endif } #endif //!DACCESS_COMPILE } // Returns the state of m_dwForbidSuspendThread as it was at the time of the call. // It may asynchronously change if there are no additional guarantees (i.e. it is the current thread or the thread is suspended) bool IsInForbidSuspendRegion() { return m_dwForbidSuspendThread.LoadWithoutBarrier() != (LONG)0; } typedef StateHolder ForbidSuspendThreadHolder; private: // Per thread counter to dispense hash code - kept in the thread so we don't need a lock // or interlocked operations to get a new hash code; DWORD m_dwHashCodeSeed; public: inline DWORD GetNewHashCode() { LIMITED_METHOD_CONTRACT; // Every thread has its own generator for hash codes so that we won't get into a situation // where two threads consistently give out the same hash codes. // Choice of multiplier guarantees period of 2**32 - see Knuth Vol 2 p16 (3.2.1.2 Theorem A). DWORD multiplier = GetThreadId()*4 + 5; m_dwHashCodeSeed = m_dwHashCodeSeed*multiplier + 1; return m_dwHashCodeSeed; } #ifdef _DEBUG // If the current thread suspends other threads, we need to make sure that the thread // only allocates memory if the suspended threads do not have OS Heap lock. static BOOL Debug_AllowCallout() { LIMITED_METHOD_CONTRACT; Thread * pThread = GetThreadNULLOk(); return ((pThread == NULL) || (pThread->dbg_m_cSuspendedThreads == pThread->dbg_m_cSuspendedThreadsWithoutOSLock)); } // Returns number of threads that are currently suspended by the current thread and that can potentially hold OS lock BOOL Debug_GetUnsafeSuspendeeCount() { LIMITED_METHOD_CONTRACT; return (dbg_m_cSuspendedThreads - dbg_m_cSuspendedThreadsWithoutOSLock); } #endif public: //-------------------------------------------------------------- // Constructor. //-------------------------------------------------------------- #ifndef DACCESS_COMPILE Thread(); #endif //-------------------------------------------------------------- // Failable initialization occurs here. //-------------------------------------------------------------- void InitThread(); BOOL AllocHandles(); //-------------------------------------------------------------- // If the thread was setup through SetupUnstartedThread, rather // than SetupThread, complete the setup here when the thread is // actually running. //-------------------------------------------------------------- BOOL HasStarted(); // We don't want ::CreateThread() calls scattered throughout the source. // Create all new threads here. The thread is created as suspended, so // you must ::ResumeThread to kick it off. It is guaranteed to create the // thread, or throw. BOOL CreateNewThread(SIZE_T stackSize, LPTHREAD_START_ROUTINE start, void *args, LPCWSTR pName=NULL); // Functions used to perform initialization and cleanup on a managed thread // that would normally occur if the thread was stated when the runtime was // fully initialized and ready to run. // Examples where this applies are the Main and Finalizer threads. static void InitializationForManagedThreadInNative(_In_ Thread* pThread); static void CleanUpForManagedThreadInNative(_In_ Thread* pThread); enum StackSizeBucket { StackSize_Small, StackSize_Medium, StackSize_Large }; // // Creates a raw OS thread; use this only for CLR-internal threads that never execute user code. // StackSizeBucket determines how large the stack should be. // static HANDLE CreateUtilityThread(StackSizeBucket stackSizeBucket, LPTHREAD_START_ROUTINE start, void *args, LPCWSTR pName, DWORD flags = 0, DWORD* pThreadId = NULL); //-------------------------------------------------------------- // Destructor //-------------------------------------------------------------- #ifndef DACCESS_COMPILE ~Thread(); #else ~Thread() {} #endif #ifdef FEATURE_COMINTEROP_APARTMENT_SUPPORT void CoUninitialize(); void BaseCoUninitialize(); #endif // FEATURE_COMINTEROP_APARTMENT_SUPPORT void CooperativeCleanup(); void OnThreadTerminate(BOOL holdingLock); static void CleanupDetachedThreads(); static void CleanupFinalizedThreads(); //-------------------------------------------------------------- // Returns innermost active Frame. //-------------------------------------------------------------- PTR_Frame GetFrame() { SUPPORTS_DAC; #ifdef _DEBUG_IMPL WRAPPER_NO_CONTRACT; if (this == GetThreadNULLOk()) { void* curSP; curSP = (void *)GetCurrentSP(); _ASSERTE(IsExecutingOnAltStack() || (curSP <= m_pFrame && m_pFrame < m_CacheStackBase) || m_pFrame == (Frame*) -1); } #endif return m_pFrame; } //-------------------------------------------------------------- // Returns innermost active GCFrame. //-------------------------------------------------------------- PTR_GCFrame GetGCFrame(); //-------------------------------------------------------------- // Replaces innermost active Frames. //-------------------------------------------------------------- #ifndef DACCESS_COMPILE void SetFrame(Frame *pFrame) #ifdef _DEBUG ; #else { LIMITED_METHOD_CONTRACT; m_pFrame = pFrame; } #endif ; #endif //-------------------------------------------------------------- // Replaces innermost active GCFrame. //-------------------------------------------------------------- #ifndef DACCESS_COMPILE void SetGCFrame(GCFrame *pFrame) { LIMITED_METHOD_CONTRACT; m_pGCFrame = pFrame; } #endif inline Frame* FindFrame(SIZE_T StackPointer); bool DetectHandleILStubsForDebugger(); //************************************************************** // GC interaction //************************************************************** //-------------------------------------------------------------- // Enter cooperative GC mode. NOT NESTABLE. //-------------------------------------------------------------- FORCEINLINE_NONDEBUG void DisablePreemptiveGC() { #ifndef DACCESS_COMPILE WRAPPER_NO_CONTRACT; _ASSERTE(this == GetThread()); _ASSERTE(!m_fPreemptiveGCDisabled); // holding a spin lock in preemp mode and transit to coop mode will cause other threads // spinning waiting for GC _ASSERTE ((m_StateNC & Thread::TSNC_OwnsSpinLock) == 0); #ifdef ENABLE_CONTRACTS_IMPL TriggersGC(this); #endif // Logically, we just want to check whether a GC is in progress and halt // at the boundary if it is -- before we disable preemptive GC. However // this opens up a race condition where the GC starts after we make the // check. SuspendRuntime will ignore such a thread because it saw it as // outside the EE. So the thread would run wild during the GC. // // Instead, enter cooperative mode and then check if a GC is in progress. // If so, go back out and try again. The reason we go back out before we // try again, is that SuspendRuntime might have seen us as being in // cooperative mode if it checks us between the next two statements. // In that case, it will be trying to move us to a safe spot. If // we don't let it see us leave, it will keep waiting on us indefinitely. // ------------------------------------------------------------------------ // ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** | // ------------------------------------------------------------------------ // // DO NOT CHANGE THIS METHOD WITHOUT VISITING ALL THE STUB GENERATORS // THAT EFFECTIVELY INLINE IT INTO THEIR STUBS // // ------------------------------------------------------------------------ // ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** | // ------------------------------------------------------------------------ m_fPreemptiveGCDisabled.StoreWithoutBarrier(1); if (g_TrapReturningThreads) { RareDisablePreemptiveGC(); } #else LIMITED_METHOD_CONTRACT; #endif } NOINLINE void RareDisablePreemptiveGC(); void HandleThreadAbort(); void PreWorkForThreadAbort(); private: void HandleThreadAbortTimeout(); public: //-------------------------------------------------------------- // Leave cooperative GC mode. NOT NESTABLE. //-------------------------------------------------------------- FORCEINLINE_NONDEBUG void EnablePreemptiveGC() { LIMITED_METHOD_CONTRACT; #ifndef DACCESS_COMPILE _ASSERTE(this == GetThread()); _ASSERTE(m_fPreemptiveGCDisabled); // holding a spin lock in coop mode and transit to preemp mode will cause deadlock on GC _ASSERTE ((m_StateNC & Thread::TSNC_OwnsSpinLock) == 0); // ------------------------------------------------------------------------ // ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** | // ------------------------------------------------------------------------ // // DO NOT CHANGE THIS METHOD WITHOUT VISITING ALL THE STUB GENERATORS // THAT EFFECTIVELY INLINE IT INTO THEIR STUBS // // ------------------------------------------------------------------------ // ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** WARNING ** | // ------------------------------------------------------------------------ m_fPreemptiveGCDisabled.StoreWithoutBarrier(0); #endif } #if defined(STRESS_HEAP) && defined(_DEBUG) void PerformPreemptiveGC(); #endif void PulseGCMode(); //-------------------------------------------------------------- // Query mode //-------------------------------------------------------------- BOOL PreemptiveGCDisabled() { WRAPPER_NO_CONTRACT; _ASSERTE(this == GetThread()); // // m_fPreemptiveGCDisabled is always modified by the thread itself, and so the thread itself // can read it without memory barrier. // return m_fPreemptiveGCDisabled.LoadWithoutBarrier(); } BOOL PreemptiveGCDisabledOther() { LIMITED_METHOD_CONTRACT; return (m_fPreemptiveGCDisabled); } #ifdef ENABLE_CONTRACTS_IMPL void BeginNoTriggerGC(const char *szFile, int lineNum) { WRAPPER_NO_CONTRACT; ClrDebugState* pClrDebugState = GetClrDebugState(); pClrDebugState->IncrementGCNoTriggerCount(); if (PreemptiveGCDisabled()) { pClrDebugState->IncrementGCForbidCount(); } } void EndNoTriggerGC() { WRAPPER_NO_CONTRACT; ClrDebugState* pClrDebugState = GetClrDebugState(); _ASSERTE(pClrDebugState->GetGCNoTriggerCount() != 0 || (pClrDebugState->ViolationMask() & BadDebugState)); pClrDebugState->DecrementGCNoTriggerCount(); if (pClrDebugState->GetGCForbidCount()) { pClrDebugState->DecrementGCForbidCount(); } } void BeginForbidGC(const char *szFile, int lineNum) { WRAPPER_NO_CONTRACT; #ifdef PROFILING_SUPPORTED _ASSERTE(PreemptiveGCDisabled() || CORProfilerPresent()); // This added to allow profiler to use GetILToNativeMapping // while in preemptive GC mode #else // PROFILING_SUPPORTED _ASSERTE(PreemptiveGCDisabled()); #endif // PROFILING_SUPPORTED BeginNoTriggerGC(szFile, lineNum); } void EndForbidGC() { WRAPPER_NO_CONTRACT; #ifdef PROFILING_SUPPORTED _ASSERTE(PreemptiveGCDisabled() || CORProfilerPresent()); // This added to allow profiler to use GetILToNativeMapping // while in preemptive GC mode #else // PROFILING_SUPPORTED _ASSERTE(PreemptiveGCDisabled()); #endif // PROFILING_SUPPORTED EndNoTriggerGC(); } BOOL GCNoTrigger() { WRAPPER_NO_CONTRACT; ClrDebugState* pClrDebugState = GetClrDebugState(); if ( (GCViolation|BadDebugState) & pClrDebugState->ViolationMask() ) { return FALSE; } return pClrDebugState->GetGCNoTriggerCount(); } BOOL GCForbidden() { WRAPPER_NO_CONTRACT; ClrDebugState* pClrDebugState = GetClrDebugState(); if ( (GCViolation|BadDebugState) & pClrDebugState->ViolationMask()) { return FALSE; } return pClrDebugState->GetGCForbidCount(); } BOOL RawGCNoTrigger() { LIMITED_METHOD_CONTRACT; ClrDebugState* pClrDebugState = GetClrDebugState(); if (pClrDebugState->ViolationMask() & BadDebugState) { return 0; } return pClrDebugState->GetGCNoTriggerCount(); } BOOL RawGCForbidden() { LIMITED_METHOD_CONTRACT; ClrDebugState* pClrDebugState = GetClrDebugState(); if (pClrDebugState->ViolationMask() & BadDebugState) { return 0; } return pClrDebugState->GetGCForbidCount(); } #endif // ENABLE_CONTRACTS_IMPL //--------------------------------------------------------------- // Calculate thread static offset //--------------------------------------------------------------- static size_t GetOffsetOfThreadStatic(void* pThreadStatic); //--------------------------------------------------------------- // Expose key offsets and values for stub generation. //--------------------------------------------------------------- static BYTE GetOffsetOfCurrentFrame() { LIMITED_METHOD_CONTRACT; size_t ofs = offsetof(class Thread, m_pFrame); _ASSERTE(FitsInI1(ofs)); return (BYTE)ofs; } static BYTE GetOffsetOfState() { LIMITED_METHOD_CONTRACT; size_t ofs = offsetof(class Thread, m_State); _ASSERTE(FitsInI1(ofs)); return (BYTE)ofs; } static BYTE GetOffsetOfGCFlag() { LIMITED_METHOD_CONTRACT; size_t ofs = offsetof(class Thread, m_fPreemptiveGCDisabled); _ASSERTE(FitsInI1(ofs)); return (BYTE)ofs; } //--------------------------------------------------------------- // Expose offset of the place for storing the filter context for the debugger. //--------------------------------------------------------------- static SIZE_T GetOffsetOfDebuggerFilterContext() { LIMITED_METHOD_CONTRACT; return (SIZE_T)(offsetof(class Thread, m_debuggerFilterContext)); } //--------------------------------------------------------------- // Expose offset of m_StateNC //--------------------------------------------------------------- static SIZE_T GetOffsetOfStateNC() { LIMITED_METHOD_CONTRACT; return (SIZE_T)(offsetof(class Thread, m_StateNC)); } //--------------------------------------------------------------- // Last exception to be thrown //--------------------------------------------------------------- OBJECTREF GetThrowable() { WRAPPER_NO_CONTRACT; return m_ExceptionState.GetThrowable(); } BOOL HasException() { LIMITED_METHOD_CONTRACT; return !IsThrowableNull(); } // See ExInfo::GetThrowableAsPseudoHandle for details on the pseudo-handle. OBJECTHANDLE GetThrowableAsPseudoHandle() { LIMITED_METHOD_DAC_CONTRACT; return m_ExceptionState.GetThrowableAsPseudoHandle(); } // special null test (for use when we're in the wrong GC mode) BOOL IsThrowableNull() { WRAPPER_NO_CONTRACT; return m_ExceptionState.IsThrowableNull(); } BOOL IsExceptionInProgress() { SUPPORTS_DAC; LIMITED_METHOD_CONTRACT; return m_ExceptionState.IsExceptionInProgress(); } void SyncManagedExceptionState(bool fIsDebuggerThread); //--------------------------------------------------------------- // Per-thread information used by handler //--------------------------------------------------------------- // exception handling info stored in thread // can't allocate this as needed because can't make exception-handling depend upon memory allocation PTR_ThreadExceptionState GetExceptionState() { LIMITED_METHOD_CONTRACT; SUPPORTS_DAC; return PTR_ThreadExceptionState(PTR_HOST_MEMBER_TADDR(Thread, this, m_ExceptionState)); } public: //--------------------------------------------------------------- // Track use of the thread block. See the general comments on // thread destruction in threads.cpp, for details. //--------------------------------------------------------------- int IncExternalCount(); int DecExternalCount(BOOL holdingLock); //--------------------------------------------------------------- // !!!! THESE ARE NOT SAFE FOR GENERAL USE !!!! // IncExternalCountDANGEROUSProfilerOnly() // DecExternalCountDANGEROUSProfilerOnly() // Currently only the profiler API should be using these // functions, because the profiler is responsible for ensuring // that the thread exists, undestroyed, before operating on it. // All other clients should use IncExternalCount/DecExternalCount // instead //--------------------------------------------------------------- int IncExternalCountDANGEROUSProfilerOnly() { LIMITED_METHOD_CONTRACT; #ifdef _DEBUG int cRefs = #else // _DEBUG return #endif //_DEBUG InterlockedIncrement((LONG*)&m_ExternalRefCount); #ifdef _DEBUG // This should never be called on a thread being destroyed _ASSERTE(cRefs != 1); return cRefs; #endif //_DEBUG } int DecExternalCountDANGEROUSProfilerOnly() { LIMITED_METHOD_CONTRACT; #ifdef _DEBUG int cRefs = #else // _DEBUG return #endif //_DEBUG InterlockedDecrement((LONG*)&m_ExternalRefCount); #ifdef _DEBUG // This should never cause the last reference on the thread to be released _ASSERTE(cRefs != 0); return cRefs; #endif //_DEBUG } // Get and Set the exposed System.Thread object which corresponds to // this thread. Also the thread handle and Id. OBJECTREF GetExposedObject(); OBJECTREF GetExposedObjectRaw(); void SetExposedObject(OBJECTREF exposed); OBJECTHANDLE GetExposedObjectHandleForDebugger() { LIMITED_METHOD_CONTRACT; return m_ExposedObject; } // Query whether the exposed object exists BOOL IsExposedObjectSet() { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_COOPERATIVE; } CONTRACTL_END; return (ObjectFromHandle(m_ExposedObject) != NULL) ; } // When we create a managed thread, the thread is suspended. We call StartThread to get // the thread start. DWORD StartThread(); // The result of attempting to OS-suspend an EE thread. enum SuspendThreadResult { // We successfully suspended the thread. This is the only // case where the caller should subsequently call ResumeThread. STR_Success, // The underlying call to the operating system's SuspendThread // or GetThreadContext failed. This is usually taken to mean // that the OS thread has exited. (This can possibly also mean // // that the suspension count exceeded the allowed maximum, but // Thread::SuspendThread asserts that does not happen.) STR_Failure, // The thread handle is invalid. This means that the thread // is dead (or dying), or that the object has been created for // an exposed System.Thread that has not been started yet. STR_UnstartedOrDead, // The fOneTryOnly flag was set, and we managed to OS suspend the // thread, but we found that it had its m_dwForbidSuspendThread // flag set. If fOneTryOnly is not set, Thread::Suspend will // retry in this case. STR_Forbidden, // Stress logging is turned on, but no stress log had been created // for the thread yet, and we failed to create one. This can mean // that either we are not allowed to call into the host, or we ran // out of memory. STR_NoStressLog, }; void Hijack(); #ifdef FEATURE_THREAD_ACTIVATION enum class ActivationReason { SuspendForGC, SuspendForDebugger, ThreadAbort, }; bool InjectActivation(ActivationReason reason); #endif // FEATURE_THREAD_ACTIVATION #ifndef DISABLE_THREADSUSPEND // SuspendThread // Attempts to OS-suspend the thread, whichever GC mode it is in. // Arguments: // fOneTryOnly - If TRUE, report failure if the thread has its // m_dwForbidSuspendThread flag set. If FALSE, retry. // pdwSuspendCount - If non-NULL, will contain the return code // of the underlying OS SuspendThread call on success, // undefined on any kind of failure. // Return value: // A SuspendThreadResult value indicating success or failure. SuspendThreadResult SuspendThread(BOOL fOneTryOnly = FALSE, DWORD *pdwSuspendCount = NULL); DWORD ResumeThread(); #endif // DISABLE_THREADSUSPEND int GetThreadPriority(); BOOL SetThreadPriority( int nPriority // thread priority level ); BOOL GetThreadContext( LPCONTEXT lpContext // context structure ) { WRAPPER_NO_CONTRACT; return ::GetThreadContext (GetThreadHandle(), lpContext); } #ifndef DACCESS_COMPILE BOOL SetThreadContext( CONST CONTEXT *lpContext // context structure ) { WRAPPER_NO_CONTRACT; return ::SetThreadContext (GetThreadHandle(), lpContext); } #endif BOOL HasValidThreadHandle () { WRAPPER_NO_CONTRACT; return GetThreadHandle() != INVALID_HANDLE_VALUE; } DWORD GetThreadId() { LIMITED_METHOD_DAC_CONTRACT; _ASSERTE(m_ThreadId != UNINITIALIZED_THREADID); return m_ThreadId; } // The actual OS thread ID may be 64 bit on some platforms but // the runtime has historically used 32 bit IDs. We continue to // downcast by default to limit the impact but GetOSThreadId64() // is available for code-paths which correctly handle it. DWORD GetOSThreadId() { LIMITED_METHOD_CONTRACT; SUPPORTS_DAC; #ifndef DACCESS_COMPILE _ASSERTE (m_OSThreadId != 0xbaadf00d); #endif // !DACCESS_COMPILE return (DWORD)m_OSThreadId; } // Allows access to the full 64 bit id on platforms which use it SIZE_T GetOSThreadId64() { LIMITED_METHOD_CONTRACT; SUPPORTS_DAC; #ifndef DACCESS_COMPILE _ASSERTE(m_OSThreadId != 0xbaadf00d); #endif // !DACCESS_COMPILE return m_OSThreadId; } // This API is to be used for Debugger only. // We need to be able to return the true value of m_OSThreadId. // On platforms with 64 bit thread IDs we downcast to 32 bit. // DWORD GetOSThreadIdForDebugger() { SUPPORTS_DAC; LIMITED_METHOD_CONTRACT; return (DWORD) m_OSThreadId; } BOOL IsThreadPoolThread() { LIMITED_METHOD_CONTRACT; return m_State & Thread::TS_TPWorkerThread; } void SetIsThreadPoolThread() { LIMITED_METHOD_CONTRACT; SetThreadState(TS_TPWorkerThread); } // public suspend functions. System ones are internal, like for GC. User ones // correspond to suspend/resume calls on the exposed System.Thread object. static bool SysStartSuspendForDebug(AppDomain *pAppDomain); static bool SysSweepThreadsForDebug(bool forceSync); static void SysResumeFromDebug(AppDomain *pAppDomain); private: // Specifies type of thread abort. DWORD m_AbortType; ULONGLONG m_AbortEndTime; ULONGLONG m_RudeAbortEndTime; BOOL m_fRudeAbortInitiated; LONG m_AbortController; static ULONGLONG s_NextSelfAbortEndTime; void SetRudeAbortEndTimeFromEEPolicy(); // This is a spin lock to serialize setting/resetting of AbortType and AbortRequest. LONG m_AbortRequestLock; static void LockAbortRequest(Thread *pThread); static void UnlockAbortRequest(Thread *pThread); typedef Holder AbortRequestLockHolder; static void AcquireAbortControl(Thread *pThread) { LIMITED_METHOD_CONTRACT; InterlockedIncrement (&pThread->m_AbortController); } static void ReleaseAbortControl(Thread *pThread) { LIMITED_METHOD_CONTRACT; _ASSERTE (pThread->m_AbortController > 0); InterlockedDecrement (&pThread->m_AbortController); } typedef Holder AbortControlHolder; public: #ifdef _DEBUG BOOL m_fRudeAborted; DWORD m_dwAbortPoint; #endif public: HRESULT UserAbort(EEPolicy::ThreadAbortTypes abortType, DWORD timeout); BOOL HandleJITCaseForAbort(); void ResetAbort(); private: void SetAbortEndTime(ULONGLONG endTime, BOOL fRudeAbort); public: ULONGLONG GetAbortEndTime() { WRAPPER_NO_CONTRACT; return IsRudeAbort()?m_RudeAbortEndTime:m_AbortEndTime; } // We distinguish interrupting a thread between Thread.Interrupt and other usage. // For Thread.Interrupt usage, we will interrupt an alertable wait using the same // rule as ReadyForAbort. Wait in EH clause or CER region is not interrupted. // For other usage, we will try to Abort the thread. // If we can not do the operation, we will delay until next wait. enum ThreadInterruptMode { TI_Interrupt = 0x00000001, // Requested by Thread.Interrupt TI_Abort = 0x00000002, // Requested by Thread.Abort or AppDomain.Unload }; private: BOOL ReadyForAsyncException(); public: BOOL ReadyForAbort() { return ReadyForAsyncException(); } BOOL IsRudeAbort(); #if defined(TARGET_AMD64) && defined(FEATURE_HIJACK) BOOL IsSafeToInjectThreadAbort(PTR_CONTEXT pContextToCheck); #endif // defined(TARGET_AMD64) && defined(FEATURE_HIJACK) inline BOOL IsAbortRequested() { LIMITED_METHOD_CONTRACT; return (m_State & TS_AbortRequested); } inline BOOL IsAbortInitiated() { LIMITED_METHOD_CONTRACT; return (m_State & TS_AbortInitiated); } inline BOOL IsRudeAbortInitiated() { LIMITED_METHOD_CONTRACT; return IsAbortRequested() && m_fRudeAbortInitiated; } inline void SetAbortInitiated() { WRAPPER_NO_CONTRACT; if (IsRudeAbort()) { m_fRudeAbortInitiated = TRUE; } SetThreadState(TS_AbortInitiated); // The following should be factored better, but I'm looking for a minimal V1 change. ResetUserInterrupted(); } inline void ResetAbortInitiated() { LIMITED_METHOD_CONTRACT; ResetThreadState(TS_AbortInitiated); m_fRudeAbortInitiated = FALSE; } inline void SetPreparingAbort() { WRAPPER_NO_CONTRACT; SetThreadStateNC(TSNC_PreparingAbort); } inline void ResetPreparingAbort() { WRAPPER_NO_CONTRACT; ResetThreadStateNC(TSNC_PreparingAbort); } private: inline static void SetPreparingAbortForHolder() { GetThread()->SetPreparingAbort(); } inline static void ResetPreparingAbortForHolder() { GetThread()->ResetPreparingAbort(); } typedef StateHolder PreparingAbortHolder; public: inline void SetIsCreatingTypeInitException() { WRAPPER_NO_CONTRACT; SetThreadStateNC(TSNC_CreatingTypeInitException); } inline void ResetIsCreatingTypeInitException() { WRAPPER_NO_CONTRACT; ResetThreadStateNC(TSNC_CreatingTypeInitException); } inline BOOL IsCreatingTypeInitException() { WRAPPER_NO_CONTRACT; return HasThreadStateNC(TSNC_CreatingTypeInitException); } private: void SetAbortRequestBit(); void RemoveAbortRequestBit(); public: void MarkThreadForAbort(EEPolicy::ThreadAbortTypes abortType); void UnmarkThreadForAbort(EEPolicy::ThreadAbortTypes abortType = EEPolicy::TA_Rude); static ULONGLONG GetNextSelfAbortEndTime() { LIMITED_METHOD_CONTRACT; return s_NextSelfAbortEndTime; } #if defined(FEATURE_HIJACK) && !defined(TARGET_UNIX) // Tricks for resuming threads from fully interruptible code with a ThreadStop. BOOL ResumeUnderControl(T_CONTEXT *pCtx); #endif // FEATURE_HIJACK && !TARGET_UNIX enum InducedThrowReason { InducedThreadStop = 1, InducedThreadRedirect = 2, InducedThreadRedirectAtEndOfCatch = 3, }; DWORD m_ThrewControlForThread; // flag that is set when the thread deliberately raises an exception for stop/abort inline DWORD ThrewControlForThread() { LIMITED_METHOD_CONTRACT; return m_ThrewControlForThread; } inline void SetThrowControlForThread(InducedThrowReason reason) { LIMITED_METHOD_CONTRACT; m_ThrewControlForThread = reason; } inline void ResetThrowControlForThread() { LIMITED_METHOD_CONTRACT; m_ThrewControlForThread = 0; } PTR_CONTEXT m_OSContext; // ptr to a Context structure used to record the OS specific ThreadContext for a thread // this is used for thread stop/abort and is initialized on demand PT_CONTEXT GetAbortContext (); // These will only ever be called from the debugger's helper // thread. // // When a thread is being created after a debug suspension has // started, we get the event on the debugger helper thread. It // will turn around and call this to set the debug suspend pending // flag on the newly created flag, since it was missed by // SysStartSuspendForGC as it didn't exist when that function was // run. void MarkForDebugSuspend(); // When the debugger uses the trace flag to single step a thread, // it also calls this function to mark this info in the thread's // state. The out-of-process portion of the debugger will read the // thread's state for a variety of reasons, including looking for // this flag. void MarkDebuggerIsStepping(bool onOff) { WRAPPER_NO_CONTRACT; if (onOff) SetThreadStateNC(Thread::TSNC_DebuggerIsStepping); else ResetThreadStateNC(Thread::TSNC_DebuggerIsStepping); } #ifdef FEATURE_EMULATE_SINGLESTEP // ARM doesn't currently support any reliable hardware mechanism for single-stepping. // ARM64 unix doesn't currently support any reliable hardware mechanism for single-stepping. // For each we emulate single step in software. This support is used only by the debugger. private: #if defined(TARGET_ARM) ArmSingleStepper m_singleStepper; #elif defined(TARGET_RISCV64) RiscV64SingleStepper m_singleStepper; #elif defined(TARGET_LOONGARCH64) LoongArch64SingleStepper m_singleStepper; #else Arm64SingleStepper m_singleStepper; #endif public: #ifndef DACCESS_COMPILE // Given the context with which this thread shall be resumed and the first WORD of the instruction that // should be executed next (this is not always the WORD under PC since the debugger uses this mechanism to // skip breakpoints written into the code), set the thread up to execute one instruction and then throw an // EXCEPTION_SINGLE_STEP. (In fact an EXCEPTION_BREAKPOINT will be thrown, but this is fixed up in our // first chance exception handler, see IsDebuggerFault in excep.cpp). void EnableSingleStep() { m_singleStepper.Enable(); } void BypassWithSingleStep(const void* ip ARM_ARG(WORD opcode1) ARM_ARG(WORD opcode2) ARM64_ARG(uint32_t opcode) RISCV64_ARG(uint32_t opcode) LOONGARCH64_ARG(uint32_t opcode)) { #if defined(TARGET_ARM) m_singleStepper.Bypass((DWORD)ip, opcode1, opcode2); #else m_singleStepper.Bypass((uint64_t)ip, opcode); #endif } void DisableSingleStep() { m_singleStepper.Disable(); } void ApplySingleStep(T_CONTEXT *pCtx) { m_singleStepper.Apply(pCtx); } bool IsSingleStepEnabled() const { return m_singleStepper.IsEnabled(); } // Fixup code called by our vectored exception handler to complete the emulation of single stepping // initiated by EnableSingleStep above. Returns true if the exception was indeed encountered during // stepping. bool HandleSingleStep(T_CONTEXT *pCtx, DWORD dwExceptionCode) { return m_singleStepper.Fixup(pCtx, dwExceptionCode); } #endif // !DACCESS_COMPILE #endif // FEATURE_EMULATE_SINGLESTEP public: // Indicate whether this thread should run in the background. Background threads // don't interfere with the EE shutting down. Whereas a running non-background // thread prevents us from shutting down (except through System.Exit(), of course) void SetBackground(BOOL isBack); #ifdef FEATURE_COMINTEROP_APARTMENT_SUPPORT // Retrieve the apartment state of the current thread. There are three possible // states: thread hosts an STA, thread is part of the MTA or thread state is // undecided. The last state may indicate that the apartment has not been set at // all (nobody has called CoInitializeEx) or that the EE does not know the // current state (EE has not called CoInitializeEx). // Keep in sync with System.Threading.ApartmentState enum ApartmentState { AS_InSTA, AS_InMTA, AS_Unknown }; ApartmentState GetApartment(); ApartmentState GetApartmentFromOS(); // Sets the apartment state if it has not already been set and // returns the state. ApartmentState GetFinalApartment(); // Attempt to set current thread's apartment state. The actual apartment state // achieved is returned and may differ from the input state if someone managed to // call CoInitializeEx on this thread first (note that calls to SetApartment made // before the thread has started are guaranteed to succeed). ApartmentState SetApartment(ApartmentState state); // Get/set apartment of a thread that was not started yet ApartmentState GetApartmentOfUnstartedThread(); void SetApartmentOfUnstartedThread(ApartmentState state); #endif // FEATURE_COMINTEROP_APARTMENT_SUPPORT // Either perform WaitForSingleObject or MsgWaitForSingleObject as appropriate. DWORD DoReentrantWaitAny(int numWaiters, HANDLE* pHandles, DWORD timeout, WaitMode mode); DWORD DoReentrantWaitWithRetry(HANDLE handle, DWORD timeout, WaitMode mode); private: DWORD DoAppropriateAptStateWait(int numWaiters, HANDLE* pHandles, BOOL bWaitAll, DWORD timeout, WaitMode mode); public: //************************************************************************ // Enumerate all frames. //************************************************************************ /* Flags used for StackWalkFramesEx */ // FUNCTIONSONLY excludes all functionless frames and all funclets #define FUNCTIONSONLY 0x0001 // SKIPFUNCLETS includes functionless frames but excludes all funclets and everything between funclets and their parent methods #define SKIPFUNCLETS 0x0002 // UNUSED 0x0004 #define QUICKUNWIND 0x0008 // do not restore all registers during unwind #define HANDLESKIPPEDFRAMES 0x0010 // temporary to handle skipped frames for appdomain unload // stack crawl. Eventually need to always do this but it // breaks the debugger right now. #define LIGHTUNWIND 0x0020 // Unwind PC+SP+FP only. // - Implemented on x64 only. // - Expects the initial context to be outside prolog/epilog. // - Cannot unwind through methods with stackalloc #define NOTIFY_ON_U2M_TRANSITIONS 0x0040 // Provide a callback for native transitions. // This is only useful to a debugger trying to find native code // in the stack. #define DISABLE_MISSING_FRAME_DETECTION 0x0080 // disable detection of missing TransitionFrames // One thread may be walking the stack of another thread // If you need to use this, you may also need to put a call to CrawlFrame::CheckGSCookies // in your callback routine if it does any potentially time-consuming activity. #define ALLOW_ASYNC_STACK_WALK 0x0100 #define THREAD_IS_SUSPENDED 0x0200 // Be careful not to cause deadlocks, this thread is suspended // Stackwalk tries to verify some objects, but it could be called in relocate phase of GC, // where objects could be in invalid state, this flag is to tell stackwalk to skip the validation #define ALLOW_INVALID_OBJECTS 0x0400 // Caller has verified that the thread to be walked is in the middle of executing // JITd or NGENd code, according to the thread's current context (or seeded // context if one was provided). The caller ensures this when the stackwalk // is initiated by a profiler. #define THREAD_EXECUTING_MANAGED_CODE 0x0800 // This stackwalk is due to the DoStackSnapshot profiler API #define PROFILER_DO_STACK_SNAPSHOT 0x1000 // When this flag is set, the stackwalker does not automatically advance to the // faulting managed stack frame when it encounters an ExInfo. This should only be // necessary for native debuggers doing mixed-mode stackwalking. #define NOTIFY_ON_NO_FRAME_TRANSITIONS 0x2000 // Normally, the stackwalker does not stop at the initial CONTEXT if the IP is in native code. // This flag changes the stackwalker behaviour. Currently this is only used in the debugger stackwalking // API. #define NOTIFY_ON_INITIAL_NATIVE_CONTEXT 0x4000 // Indicates that we are enumerating GC references and should follow appropriate // callback rules for parent methods vs funclets. Only supported on non-x86 platforms. // // Refer to StackFrameIterator::Filter for detailed comments on this flag. #define GC_FUNCLET_REFERENCE_REPORTING 0x8000 // Stackwalking normally checks GS cookies on the fly, but there are cases in which the JIT reports // incorrect epilog information. This causes the debugger to request stack walks in the epilog, checking // an now invalid cookie. This flag allows the debugger stack walks to disable GS cookie checking. // This is a workaround for the debugger stackwalking. In general, the stackwalker and CrawlFrame // may still execute GS cookie tracking/checking code paths. #define SKIP_GSCOOKIE_CHECK 0x10000 #define UNWIND_FLOATS 0x20000 StackWalkAction StackWalkFramesEx( PREGDISPLAY pRD, // virtual register set at crawl start PSTACKWALKFRAMESCALLBACK pCallback, VOID *pData, unsigned flags, PTR_Frame pStartFrame = PTR_NULL); private: // private helpers used by StackWalkFramesEx and StackFrameIterator StackWalkAction MakeStackwalkerCallback(CrawlFrame* pCF, PSTACKWALKFRAMESCALLBACK pCallback, VOID* pData DEBUG_ARG(UINT32 uLoopIteration)); #ifdef _DEBUG void DebugLogStackWalkInfo(CrawlFrame* pCF, _In_z_ LPCSTR pszTag, UINT32 uLoopIteration); #endif // _DEBUG public: StackWalkAction StackWalkFrames( PSTACKWALKFRAMESCALLBACK pCallback, VOID *pData, unsigned flags = 0, PTR_Frame pStartFrame = PTR_NULL); bool InitRegDisplay(const PREGDISPLAY, const PT_CONTEXT, bool validContext); void FillRegDisplay(const PREGDISPLAY pRD, PT_CONTEXT pctx, bool fLightUnwind = false); static PCODE VirtualUnwindCallFrame(T_CONTEXT* pContext, T_KNONVOLATILE_CONTEXT_POINTERS* pContextPointers = NULL, EECodeInfo * pCodeInfo = NULL ARM64_ARG(TADDR * pSpForPacSign = NULL)); static UINT_PTR VirtualUnwindCallFrame(PREGDISPLAY pRD, EECodeInfo * pCodeInfo = NULL); #ifndef DACCESS_COMPILE static PCODE VirtualUnwindLeafCallFrame(T_CONTEXT* pContext); static UINT_PTR VirtualUnwindToFirstManagedCallFrame(T_CONTEXT* pContext); #endif // DACCESS_COMPILE // During a , this thread must not be asynchronously // stopped or interrupted. That would leave the class unavailable // and is therefore a security hole. static void IncPreventAsync() { WRAPPER_NO_CONTRACT; Thread *pThread = GetThread(); InterlockedIncrement((LONG*)&pThread->m_PreventAsync); } static void DecPreventAsync() { WRAPPER_NO_CONTRACT; Thread *pThread = GetThread(); InterlockedDecrement((LONG*)&pThread->m_PreventAsync); } bool IsAsyncPrevented() { return m_PreventAsync != 0; } typedef StateHolder ThreadPreventAsyncHolder; // While executing the new exception handling managed code, // this thread must not be aborted. static void IncPreventAbort() { WRAPPER_NO_CONTRACT; Thread *pThread = GetThread(); InterlockedIncrement((LONG*)&pThread->m_PreventAbort); } static void DecPreventAbort() { WRAPPER_NO_CONTRACT; Thread *pThread = GetThread(); #ifdef _DEBUG LONG c = #endif // _DEBUG InterlockedDecrement((LONG*)&pThread->m_PreventAbort); _ASSERTE(c >= 0); } BOOL IsAbortPrevented() { return m_PreventAbort != 0; } // The ThreadStore manages a list of all the threads in the system. // Next pointer for SList linkage (ThreadStore::m_ThreadList). PTR_Thread m_pNext = NULL; // Debugger per-thread flag for enabling notification on "manual" // method calls, for stepping logic void IncrementTraceCallCount(); void DecrementTraceCallCount(); FORCEINLINE int IsTraceCall() { LIMITED_METHOD_CONTRACT; return m_TraceCallCount; } private: #if defined(FEATURE_HIJACK) && !defined(TARGET_UNIX) // Used in suspension code to redirect a thread at a HandledJITCase BOOL RedirectThreadAtHandledJITCase(PFN_REDIRECTTARGET pTgt); BOOL RedirectCurrentThreadAtHandledJITCase(PFN_REDIRECTTARGET pTgt, T_CONTEXT *pCurrentThreadCtx); // Will Redirect the thread using RedirectThreadAtHandledJITCase if necessary BOOL CheckForAndDoRedirect(PFN_REDIRECTTARGET pRedirectTarget); BOOL CheckForAndDoRedirectForDbg(); BOOL CheckForAndDoRedirectForGC(); BOOL CheckForAndDoRedirectForUserSuspend(); // Exception handling must be very aware of redirection, so we provide a helper // to identifying redirection targets static BOOL IsAddrOfRedirectFunc(void * pFuncAddr); #if defined(HAVE_GCCOVER) && defined(USE_REDIRECT_FOR_GCSTRESS) public: BOOL CheckForAndDoRedirectForGCStress (T_CONTEXT *pCurrentThreadCtx); #endif // HAVE_GCCOVER && USE_REDIRECT_FOR_GCSTRESS #endif // FEATURE_HIJACK && !TARGET_UNIX public: #ifndef DACCESS_COMPILE // These re-calculate the proper value on each call for the currently executing thread. Use GetCachedStackLimit // and GetCachedStackBase for the cached values on this Thread. static void * GetStackLowerBound(); static void * GetStackUpperBound(); bool CheckCanUseStackAlloc() { int local; UINT_PTR current = reinterpret_cast(&local); UINT_PTR limit = GetCachedStackStackAllocNonRiskyExecutionLimit(); return (current > limit); } #else // DACCESS_COMPILE bool CheckCanUseStackAlloc() { return true; } #endif enum SetStackLimitScope { fAll, fAllowableOnly }; BOOL SetStackLimits(SetStackLimitScope scope); // These access the stack base and limit values for this thread. (They are cached during InitThread.) The // "stack base" is the "upper bound", i.e., where the stack starts growing from. (Main's call frame is at the // upper bound.) The "stack limit" is the "lower bound", i.e., how far the stack can grow down to. // The "stack sufficient execution limit" is used by TryEnsureSufficientExecutionStack() to limit how much stack // should remain to execute the average Framework method. PTR_VOID GetCachedStackBase() {LIMITED_METHOD_DAC_CONTRACT; return m_CacheStackBase; } PTR_VOID GetCachedStackLimit() {LIMITED_METHOD_DAC_CONTRACT; return m_CacheStackLimit;} UINT_PTR GetCachedStackSufficientExecutionLimit() {LIMITED_METHOD_DAC_CONTRACT; return m_CacheStackSufficientExecutionLimit;} UINT_PTR GetCachedStackStackAllocNonRiskyExecutionLimit() {LIMITED_METHOD_DAC_CONTRACT; return m_CacheStackStackAllocNonRiskyExecutionLimit;} private: // Access the base and limit of the stack. (I.e. the memory ranges that the thread has reserved for its stack). // // Note that the base is at a higher address than the limit, since the stack grows downwards. // // Note that we generally access the stack of the thread we are crawling, which is cached in the ScanContext. PTR_VOID m_CacheStackBase; PTR_VOID m_CacheStackLimit; UINT_PTR m_CacheStackSufficientExecutionLimit; UINT_PTR m_CacheStackStackAllocNonRiskyExecutionLimit; private: // static HRESULT CLRSetThreadStackGuarantee(SetThreadStackGuaranteeScope fScope = STSGuarantee_OnlyIfEnabled); // try to turn a page into a guard page static BOOL MarkPageAsGuard(UINT_PTR uGuardPageBase); // scan a region for a guard page static BOOL DoesRegionContainGuardPage(UINT_PTR uLowAddress, UINT_PTR uHighAddress); // Every stack has a single reserved page at its limit that we call the 'hard guard page'. This page is never // committed, and access to it after a stack overflow will terminate the thread. #define HARD_GUARD_REGION_SIZE (minipal_getpagesize()) #define SIZEOF_DEFAULT_STACK_GUARANTEE (minipal_getpagesize()) public: // This will return the last stack address that one could write to before a stack overflow. static UINT_PTR GetLastNormalStackAddress(UINT_PTR stackBase); UINT_PTR GetLastNormalStackAddress(); void ResetStackLimits() { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_ANY; } CONTRACTL_END; if (!IsSetThreadStackGuaranteeInUse()) { return; } SetStackLimits(fAllowableOnly); } INDEBUG(static void DebugLogStackMBIs()); #if defined(_DEBUG_IMPL) && !defined(DACCESS_COMPILE) // Verify that the cached stack base is for the current thread. BOOL HasRightCacheStackBase() { WRAPPER_NO_CONTRACT; return m_CacheStackBase == GetStackUpperBound(); } #endif public: static BOOL UniqueStack(void* startLoc = 0); BOOL IsAddressInStack (PTR_VOID addr) const; static BOOL IsAddressInCurrentStack (PTR_VOID addr); // DetermineIfGuardPagePresent returns TRUE if the thread's stack contains a proper guard page. This function // makes a physical check of the stack, rather than relying on whether or not the CLR is currently processing a // stack overflow exception. BOOL DetermineIfGuardPagePresent(); // Returns the amount of stack available after an SO but before the OS rips the process. static UINT_PTR GetStackGuarantee(); // RestoreGuardPage will replace the guard page on this thread's stack. The assumption is that it was removed // by the OS due to a stack overflow exception. This function requires that you know that you have enough stack // space to restore the guard page, so make sure you know what you're doing when you decide to call this. VOID RestoreGuardPage(); #if defined(FEATURE_HIJACK) && !defined(TARGET_UNIX) private: // Redirecting of threads in managed code at suspension enum RedirectReason { RedirectReason_GCSuspension, RedirectReason_DebugSuspension, RedirectReason_UserSuspension, #if defined(HAVE_GCCOVER) && defined(USE_REDIRECT_FOR_GCSTRESS) // GCCOVER RedirectReason_GCStress, #endif // HAVE_GCCOVER && USE_REDIRECT_FOR_GCSTRESS }; static void __stdcall RedirectedHandledJITCase(RedirectReason reason); static void __stdcall RedirectedHandledJITCaseForDbgThreadControl(); static void __stdcall RedirectedHandledJITCaseForGCThreadControl(); static void __stdcall RedirectedHandledJITCaseForUserSuspend(); #if defined(HAVE_GCCOVER) && defined(USE_REDIRECT_FOR_GCSTRESS) // GCCOVER static void __stdcall RedirectedHandledJITCaseForGCStress(); #endif // defined(HAVE_GCCOVER) && USE_REDIRECT_FOR_GCSTRESS #ifdef TARGET_X86 // RtlRestoreContext is available on x86, but relatively recently. // RestoreContextSimulated uses SEH machinery for a similar result on legacy OS-es. // This function should not be used on new OS-es as the pattern is not // guaranteed to continue working in the future. static void RestoreContextSimulated(Thread* pThread, CONTEXT* pCtx, void* pFrame, DWORD dwLastError); #endif friend void CPFH_AdjustContextForThreadSuspensionRace(T_CONTEXT *pContext, Thread *pThread); #endif // FEATURE_HIJACK && !TARGET_UNIX private: //------------------------------------------------------------- // Waiting & Synchronization //------------------------------------------------------------- // For suspends. The thread waits on this event. A client sets the event to cause // the thread to resume. void WaitForDebugSuspend(); BOOL WaitForDebugSuspendHelper(void); // Helpers to ensure that the bits for suspension and the number of active // traps remain coordinated. void MarkForSuspension(ULONG bit); void UnmarkForSuspension(ULONG bit); void SetupForSuspension(ULONG bit) { WRAPPER_NO_CONTRACT; if (bit & TS_DebugSuspendPending) { m_DebugSuspendEvent.Reset(); } } void ReleaseFromSuspension(ULONG bit) { WRAPPER_NO_CONTRACT; UnmarkForSuspension(~bit); // // If the thread is set free, mark it as not-suspended now // ThreadState oldState = m_State; while ((oldState & TS_DebugSuspendPending) == 0) { // // Construct the destination state we desire - all suspension bits turned off. // ThreadState newState = (ThreadState)(oldState & ~(TS_DebugSuspendPending | TS_DebugSyncSuspended)); if (InterlockedCompareExchange((LONG *)&m_State, newState, oldState) == (LONG)oldState) { break; } // // The state changed underneath us, refresh it and try again. // oldState = m_State; } if (bit & TS_DebugSuspendPending) { m_DebugSuspendEvent.Set(); } } public: FORCEINLINE void UnhijackThreadNoAlloc() { #if defined(FEATURE_HIJACK) && !defined(DACCESS_COMPILE) if (m_State & TS_Hijacked) { *m_ppvHJRetAddrPtr = m_pvHJRetAddr; ResetThreadState(TS_Hijacked); } #endif } void UnhijackThread(); // Flags that may be passed to GetSafelyRedirectableThreadContext, to customize // which checks it should perform. This allows a subset of the context verification // logic used by HandledJITCase to be shared with other callers, such as profiler // stackwalking enum GetSafelyRedirectableThreadContextOptions { // Perform the default thread context checks kDefaultChecks = 0x00000000, // Compares the thread context's IP against m_LastRedirectIP, and potentially // updates m_LastRedirectIP, when determining the safeness of the thread's // context. HandledJITCase will always set this flag. // This flag is ignored on non-x86 platforms, and also on x86 if the OS supports // trap frame reporting. kPerfomLastRedirectIPCheck = 0x00000001, // Use g_pDebugInterface->IsThreadContextInvalid() to see if breakpoints might // confuse the stack walker. HandledJITCase will always set this flag. kCheckDebuggerBreakpoints = 0x00000002, }; // Helper used by HandledJITCase and others who need an absolutely reliable // register context. BOOL GetSafelyRedirectableThreadContext(DWORD dwOptions, T_CONTEXT * pCtx, REGDISPLAY * pRD); private: #ifdef FEATURE_HIJACK void HijackThread(ExecutionState *esb X86_ARG(ReturnKind returnKind) X86_ARG(bool hasAsyncRet)); VOID *m_pvHJRetAddr; // original return address (before hijack) #ifdef TARGET_ARM64 VOID *m_pSpForPacSign; // stack pointer value that was used to sign LR with PACIASP #endif VOID **m_ppvHJRetAddrPtr; // place we bashed a new return address MethodDesc *m_HijackedFunction; // remember what we hijacked #ifndef TARGET_UNIX BOOL HandledJITCase(); #ifdef TARGET_X86 PCODE m_LastRedirectIP; ULONG m_SpinCount; #endif // TARGET_X86 #endif // !TARGET_UNIX #endif // FEATURE_HIJACK // support for Thread.Interrupt() which breaks out of Waits, Sleeps, Joins LONG m_UserInterrupt; DWORD IsUserInterrupted() { LIMITED_METHOD_CONTRACT; return m_UserInterrupt; } void ResetUserInterrupted() { LIMITED_METHOD_CONTRACT; InterlockedExchange(&m_UserInterrupt, 0); } #ifdef TARGET_WINDOWS static void WINAPI UserInterruptAPC(ULONG_PTR ignore); public: void UserInterrupt(ThreadInterruptMode mode); #endif // TARGET_WINDOWS public: void HandleThreadInterrupt(); // Access to thread handle and ThreadId. HANDLE GetThreadHandle() { LIMITED_METHOD_CONTRACT; #if defined(_DEBUG) && !defined(DACCESS_COMPILE) { CounterHolder handleHolder(&m_dwThreadHandleBeingUsed); HANDLE handle = m_ThreadHandle; _ASSERTE ( handle == INVALID_HANDLE_VALUE || m_OSThreadId == 0 || m_OSThreadId == 0xbaadf00d || ::MatchThreadHandleToOsId(handle, (DWORD)m_OSThreadId) ); } #endif DACCOP_IGNORE(FieldAccess, "Treated as raw address, no marshaling is necessary"); return m_ThreadHandle; } private: // For suspends: CLREvent m_DebugSuspendEvent; void SetThreadHandle(HANDLE h) { LIMITED_METHOD_CONTRACT; #if defined(_DEBUG) _ASSERTE ( h == INVALID_HANDLE_VALUE || m_OSThreadId == 0 || m_OSThreadId == 0xbaadf00d || ::MatchThreadHandleToOsId(h, (DWORD)m_OSThreadId) ); #endif InterlockedExchangeT(&m_ThreadHandle, h); } // We maintain a correspondence between this object, the ThreadId and ThreadHandle // in Win32, and the exposed Thread object. HANDLE m_ThreadHandle; // It would be nice to remove m_ThreadHandleForClose to simplify Thread.Join, // but at the moment that isn't possible without extensive work. // This handle is used by SwitchOut to store the old handle that needs to be closed. // The handle can't be closed before checking the external count, // which we can't do in SwitchOut since that may require locking or switching threads. HANDLE m_ThreadHandleForClose; HANDLE m_ThreadHandleForResume; SIZE_T m_OSThreadId; BOOL CreateNewOSThread(SIZE_T stackSize, LPTHREAD_START_ROUTINE start, void *args); OBJECTHANDLE m_ExposedObject; OBJECTHANDLE m_StrongHndToExposedObject; DWORD m_Priority; // initialized to INVALID_THREAD_PRIORITY, set to actual priority when a // thread does a busy wait for GC, reset to INVALID_THREAD_PRIORITY after wait is over #ifdef HAVE_GCCOVER friend void DoGcStress (PT_CONTEXT regs, NativeCodeVersion nativeCodeVersion); // Needs to call UnhijackThread #endif // HAVE_GCCOVER ULONG m_ExternalRefCount; LONG m_TraceCallCount; private: //----------------------------------------------------------- // Last exception to be thrown. //----------------------------------------------------------- friend class EEDbgInterfaceImpl; private: // Stores the most recently thrown exception. We need to have a handle in case a GC occurs before // we catch so we don't lose the object. Having a static allows others to catch outside of CLR w/o leaking // a handler and allows rethrow outside of CLR too. // Differs from m_pThrowable in that it doesn't stack on nested exceptions. OBJECTHANDLE m_LastThrownObjectHandle; // Unsafe to use directly. Use accessors instead. // Indicates that the throwable in m_lastThrownObjectHandle should be treated as // unhandled. This occurs during fatal error and a few other early error conditions // before EH is fully set up. BOOL m_ltoIsUnhandled; friend void DECLSPEC_NORETURN EEPolicy::HandleFatalStackOverflow(EXCEPTION_POINTERS *pExceptionInfo, BOOL fSkipDebugger); public: BOOL IsLastThrownObjectNull() { WRAPPER_NO_CONTRACT; return (m_LastThrownObjectHandle == (OBJECTHANDLE)0); } OBJECTREF LastThrownObject() { WRAPPER_NO_CONTRACT; if (m_LastThrownObjectHandle == (OBJECTHANDLE)0) { return NULL; } else { // We only have a handle if we have an object to keep in it. _ASSERTE(ObjectFromHandle(m_LastThrownObjectHandle) != NULL); return ObjectFromHandle(m_LastThrownObjectHandle); } } OBJECTHANDLE LastThrownObjectHandle() { LIMITED_METHOD_DAC_CONTRACT; return m_LastThrownObjectHandle; } void SetLastThrownObject(OBJECTREF throwable, BOOL isUnhandled = FALSE); void SetSOForLastThrownObject(); OBJECTREF SafeSetLastThrownObject(OBJECTREF throwable); // Inidcates that the last thrown object is now treated as unhandled void MarkLastThrownObjectUnhandled() { LIMITED_METHOD_CONTRACT; m_ltoIsUnhandled = TRUE; } // TRUE if the throwable in LTO should be treated as unhandled BOOL IsLastThrownObjectUnhandled() { LIMITED_METHOD_DAC_CONTRACT; return m_ltoIsUnhandled; } void SafeUpdateLastThrownObject(void); OBJECTREF SafeSetThrowables(OBJECTREF pThrowable, BOOL isUnhandled = FALSE); bool IsLastThrownObjectStackOverflowException() { LIMITED_METHOD_CONTRACT; CONSISTENCY_CHECK(NULL != g_pPreallocatedStackOverflowException); return (m_LastThrownObjectHandle == g_pPreallocatedStackOverflowException); } // get the current notification (if any) from this thread OBJECTHANDLE GetThreadCurrNotification(); // set the current notification on this thread void SetThreadCurrNotification(OBJECTHANDLE handle); // clear the current notification (if any) from this thread void ClearThreadCurrNotification(); private: void SetLastThrownObjectHandle(OBJECTHANDLE h); ThreadExceptionState m_ExceptionState; private: //--------------------------------------------------------------- // m_debuggerFilterContext holds the thread's "filter context" for the // debugger. This filter context is used by the debugger to seed // stack walks on the thread. //--------------------------------------------------------------- PTR_CONTEXT m_debuggerFilterContext; //--------------------------------------------------------------- // m_hijackLock holds a BOOL that is used for mutual exclusion // between profiler stack walks and thread hijacks (bashing // return addresses on the stack) //--------------------------------------------------------------- Volatile m_hijackLock; //--------------------------------------------------------------- // The current custom notification data object (or NULL if none // pending) //--------------------------------------------------------------- OBJECTHANDLE m_hCurrNotification; //--------------------------------------------------------------- // For Interop-Debugging; track if a thread is hijacked. //--------------------------------------------------------------- BOOL m_fInteropDebuggingHijacked; #if defined(PROFILING_SUPPORTED) || defined(PROFILING_SUPPORTED_DATA) //--------------------------------------------------------------- // m_profilerFilterContext holds an additional context for the // case when a (sampling) profiler wishes to hijack the thread // and do a stack walk on the same thread. //--------------------------------------------------------------- T_CONTEXT *m_pProfilerFilterContext; //--------------------------------------------------------------- // Bitmask to remember per-thread state useful for the profiler API. See // COR_PRF_CALLBACKSTATE_* flags in clr\src\inc\ProfilePriv.h for bit values. //--------------------------------------------------------------- DWORD m_profilerCallbackState; //--------------------------------------------------------------- // m_dwProfilerEvacuationCounters keeps track of how many profiler // callback calls remain on the stack //--------------------------------------------------------------- // Why volatile? // See code:ProfilingAPIUtility::InitializeProfiling#LoadUnloadCallbackSynchronization. Volatile m_dwProfilerEvacuationCounters[MAX_NOTIFICATION_PROFILERS + 1]; #endif // PROFILING_SUPPORTED || PROFILING_SUPPORTED_DATA private: #ifndef DACCESS_COMPILE private: static UINT32 *GetThreadLocalCountRef(Thread *pThread, SIZE_T threadLocalCountOffset) { WRAPPER_NO_CONTRACT; _ASSERTE(threadLocalCountOffset <= sizeof(Thread) - sizeof(UINT32)); return (UINT32 *)((SIZE_T)pThread + threadLocalCountOffset); } static void IncrementCount(Thread *pThread, SIZE_T threadLocalCountOffset, UINT64 *overflowCount) { WRAPPER_NO_CONTRACT; _ASSERTE(overflowCount != nullptr); if (pThread != nullptr) { UINT32 *threadLocalCount = GetThreadLocalCountRef(pThread, threadLocalCountOffset); UINT32 newCount = *threadLocalCount + 1; if (newCount != 0) { VolatileStoreWithoutBarrier(threadLocalCount, newCount); } else { OnIncrementCountOverflow(threadLocalCount, overflowCount); } } else { InterlockedIncrement64((LONGLONG *)overflowCount); } } static void OnIncrementCountOverflow(UINT32 *threadLocalCount, UINT64 *overflowCount); static UINT64 GetOverflowCount(UINT64 *overflowCount) { WRAPPER_NO_CONTRACT; if (sizeof(void *) >= sizeof(*overflowCount)) { return VolatileLoad(overflowCount); } return InterlockedCompareExchange64((LONGLONG *)overflowCount, 0, 0); // prevent tearing } static UINT64 GetTotalCount(SIZE_T threadLocalCountOffset, UINT64 *overflowCount); #endif // !DACCESS_COMPILE public: void SetInteropDebuggingHijacked(BOOL f) { LIMITED_METHOD_CONTRACT; m_fInteropDebuggingHijacked = f; } BOOL GetInteropDebuggingHijacked() { LIMITED_METHOD_CONTRACT; return m_fInteropDebuggingHijacked; } void SetFilterContext(T_CONTEXT *pContext); T_CONTEXT *GetFilterContext(void); #if defined(PROFILING_SUPPORTED) || defined(PROFILING_SUPPORTED_DATA) void SetProfilerFilterContext(T_CONTEXT *pContext) { LIMITED_METHOD_CONTRACT; m_pProfilerFilterContext = pContext; } FORCEINLINE DWORD GetProfilerEvacuationCounter(size_t slot) { LIMITED_METHOD_CONTRACT; _ASSERTE(slot >= 0 && slot <= MAX_NOTIFICATION_PROFILERS); return m_dwProfilerEvacuationCounters[slot].Load(); } FORCEINLINE void IncProfilerEvacuationCounter(size_t slot) { // All manipulation of the evacuation counters must be done from within the thread. A value of 0 or non-zero signals to other threads that various behavior should occur. LIMITED_METHOD_CONTRACT; _ASSERTE(slot >= 0 && slot <= MAX_NOTIFICATION_PROFILERS); #ifdef _DEBUG DWORD newValue = #endif // _DEBUG m_dwProfilerEvacuationCounters[slot] = m_dwProfilerEvacuationCounters[slot].Load() + 1; _ASSERTE(newValue != 0U); } FORCEINLINE void DecProfilerEvacuationCounter(size_t slot) { LIMITED_METHOD_CONTRACT; // All manipulation of the evacuation counters must be done from within the thread. A value of 0 or non-zero signals to other threads that various behavior should occur. _ASSERTE(slot >= 0 && slot <= MAX_NOTIFICATION_PROFILERS); #ifdef _DEBUG DWORD newValue = #endif // _DEBUG m_dwProfilerEvacuationCounters[slot] = m_dwProfilerEvacuationCounters[slot].Load() - 1; _ASSERTE(newValue != (DWORD)-1); } // Used by the profiler API to find which flags have been set on the Thread object, // in order to authorize a profiler's call into ICorProfilerInfo(2). DWORD GetProfilerCallbackFullState() { LIMITED_METHOD_CONTRACT; _ASSERTE(GetThread() == this); return m_profilerCallbackState; } // Used by profiler API to set at once all callback flag bits stored on the Thread object. // Used to reinstate the previous state that had been modified by a previous call to // SetProfilerCallbackStateFlags void SetProfilerCallbackFullState(DWORD dwFullState) { LIMITED_METHOD_CONTRACT; _ASSERTE(GetThread() == this); m_profilerCallbackState = dwFullState; } // Used by profiler API to set individual callback flags on the Thread object. // Returns the previous state of all flags. DWORD SetProfilerCallbackStateFlags(DWORD dwFlags) { LIMITED_METHOD_CONTRACT; _ASSERTE(GetThread() == this); DWORD dwRet = m_profilerCallbackState; m_profilerCallbackState |= dwFlags; return dwRet; } #endif // PROFILING_SUPPORTED || PROFILING_SUPPORTED_DATA T_CONTEXT *GetProfilerFilterContext(void) { LIMITED_METHOD_CONTRACT; #if defined(PROFILING_SUPPORTED) || defined(PROFILING_SUPPORTED_DATA) return m_pProfilerFilterContext; #else return NULL; #endif // PROFILING_SUPPORTED || PROFILING_SUPPORTED_DATA } //------------------------------------------------------------------------- // The hijack lock enforces that a thread on which a profiler is currently // performing a stack walk cannot be hijacked. // // Note that the hijack lock cannot be managed by the host (i.e., this // cannot be a Crst), because this could lead to a deadlock: YieldTask, // which is called by the host, may need to hijack, for which it would // need to take this lock - but since the host needs not be reentrant, // taking the lock cannot cause a call back into the host. //------------------------------------------------------------------------- static BOOL EnterHijackLock(Thread *pThread) { LIMITED_METHOD_CONTRACT; return ::InterlockedCompareExchange(&(pThread->m_hijackLock), TRUE, FALSE) == FALSE; } static void LeaveHijackLock(Thread *pThread) { LIMITED_METHOD_CONTRACT; pThread->m_hijackLock = FALSE; } typedef ConditionalStateHolder HijackLockHolder; //------------------------------------------------------------------------- void PrepareForEERestart(BOOL SuspendSucceeded) { WRAPPER_NO_CONTRACT; #ifdef FEATURE_HIJACK // Only unhijack the thread if the suspend succeeded. If it failed, // the target thread may currently be using the original stack // location of the return address for something else. if (SuspendSucceeded) UnhijackThread(); #endif // FEATURE_HIJACK } static LPVOID GetStaticFieldAddress(FieldDesc *pFD); TADDR GetStaticFieldAddrNoCreate(FieldDesc *pFD); private: // Don't allow a thread to be asynchronously stopped or interrupted (e.g. because // it is performing a ) int m_PreventAsync; // Don't allow a thread to be aborted while running the new exception handling managed code int m_PreventAbort; static LONG m_DebugWillSyncCount; #define BEGIN_FORBID_TYPELOAD() _ASSERTE_IMPL((GetThreadNULLOk() == 0) || ++GetThreadNULLOk()->m_ulForbidTypeLoad) #define END_FORBID_TYPELOAD() _ASSERTE_IMPL((GetThreadNULLOk() == 0) || GetThreadNULLOk()->m_ulForbidTypeLoad--) #define TRIGGERS_TYPELOAD() _ASSERTE_IMPL((GetThreadNULLOk() == 0) || !GetThreadNULLOk()->m_ulForbidTypeLoad) #ifdef _DEBUG public: DWORD m_GCOnTransitionsOK; ULONG m_ulForbidTypeLoad; /****************************************************************************/ /* The code below an attempt to catch people who don't protect GC pointers that they should be protecting. Basically, OBJECTREF's constructor, adds the slot to a table. When we protect a slot, we remove it from the table. When GC could happen, all entries in the table are marked as bad. When access to an OBJECTREF happens (the -> operator) we assert the slot is not bad. To make this fast, the table is not perfect (there can be collisions), but this should not cause false positives, but it may allow errors to go undetected */ #ifdef HOST_64BIT #define OBJREF_HASH_SHIFT_AMOUNT 3 #else // HOST_64BIT #define OBJREF_HASH_SHIFT_AMOUNT 2 #endif // HOST_64BIT // For debugging, you may want to make this number very large, (8K) // should basically ensure that no collisions happen #define OBJREF_TABSIZE 256 DWORD_PTR dangerousObjRefs[OBJREF_TABSIZE]; // Really objectRefs with lower bit stolen // m_allObjRefEntriesBad is TRUE iff dangerousObjRefs are all marked as GC happened // It's purely a perf optimization for debug builds that'll help for the cases where we make 2 successive calls // to Thread::TriggersGC. In that case, the entire array doesn't need to be walked and marked, since we just did // that. BOOL m_allObjRefEntriesBad; static DWORD_PTR OBJREF_HASH; // Remembers that this object ref pointer is 'alive' and unprotected (Bad if GC happens) static void ObjectRefNew(const OBJECTREF* ref) { WRAPPER_NO_CONTRACT; Thread * curThread = GetThreadNULLOk(); if (curThread == 0) return; curThread->dangerousObjRefs[((size_t)ref >> OBJREF_HASH_SHIFT_AMOUNT) % OBJREF_HASH] = (size_t)ref; curThread->m_allObjRefEntriesBad = FALSE; } static void ObjectRefAssign(const OBJECTREF* ref) { WRAPPER_NO_CONTRACT; Thread * curThread = GetThreadNULLOk(); if (curThread == 0) return; curThread->m_allObjRefEntriesBad = FALSE; DWORD_PTR* slot = &curThread->dangerousObjRefs[((DWORD_PTR) ref >> OBJREF_HASH_SHIFT_AMOUNT) % OBJREF_HASH]; if ((*slot & ~3) == (size_t) ref) *slot = *slot & ~1; // Don't care about GC's that have happened } // If an object is protected, it can be removed from the 'dangerous table' static void ObjectRefProtected(const OBJECTREF* ref) { #ifdef USE_CHECKED_OBJECTREFS WRAPPER_NO_CONTRACT; _ASSERTE(IsObjRefValid(ref)); Thread * curThread = GetThreadNULLOk(); if (curThread == 0) return; curThread->m_allObjRefEntriesBad = FALSE; DWORD_PTR* slot = &curThread->dangerousObjRefs[((DWORD_PTR) ref >> OBJREF_HASH_SHIFT_AMOUNT) % OBJREF_HASH]; if ((*slot & ~3) == (DWORD_PTR) ref) *slot = (size_t) ref | 2; // mark has being protected #else LIMITED_METHOD_CONTRACT; #endif } static bool IsObjRefValid(const OBJECTREF* ref) { WRAPPER_NO_CONTRACT; Thread * curThread = GetThreadNULLOk(); if (curThread == 0) return(true); // If the object ref is NULL, we'll let it pass. if (*((DWORD_PTR*) ref) == 0) return(true); DWORD_PTR val = curThread->dangerousObjRefs[((DWORD_PTR) ref >> OBJREF_HASH_SHIFT_AMOUNT) % OBJREF_HASH]; // if not in the table, or not the case that it was unprotected and GC happened, return true. if((val & ~3) != (size_t) ref || (val & 3) != 1) return(true); // If the pointer lives in the GC heap, than it is protected, and thus valid. if (dac_cast(g_lowest_address) <= val && val < dac_cast(g_highest_address)) return(true); // Same for frozen segments if (GCHeapUtilities::GetGCHeap()->IsInFrozenSegment(*(Object**)ref)) return(true); return(false); } // Clears the table. Useful to do when crossing the managed-code - EE boundary // as you usually only care about OBJECTREFS that have been created after that static void STDCALL ObjectRefFlush(Thread* thread); #ifdef ENABLE_CONTRACTS_IMPL // Marks all Objrefs in the table as bad (since they are unprotected) static void TriggersGC(Thread* thread) { WRAPPER_NO_CONTRACT; if ((GCViolation|BadDebugState) & (UINT_PTR)(GetViolationMask())) { return; } if (!thread->m_allObjRefEntriesBad) { thread->m_allObjRefEntriesBad = TRUE; for(unsigned i = 0; i < OBJREF_TABSIZE; i++) thread->dangerousObjRefs[i] |= 1; // mark all slots as GC happened } } #endif // ENABLE_CONTRACTS_IMPL #endif // _DEBUG private: // context used during redirection of this thread // NOTE: there is only one. Since redirection cannot be nested // if more than one are needed, something is wrong. PTR_CONTEXT m_pSavedRedirectContext; // in a case when we need the redirection context to include CONTEXT_XSTATE // this is the buffer that contains the context parts. // we need the buffer so we could deallocate the whole deal. BYTE* m_pOSContextBuffer; #ifdef _DEBUG // validate that we use only one context per thread. bool m_RedirectContextInUse; #endif BOOL IsContextSafeToRedirect(const T_CONTEXT* pContext); public: PT_CONTEXT GetSavedRedirectContext() { LIMITED_METHOD_CONTRACT; return (m_pSavedRedirectContext); } #ifndef DACCESS_COMPILE void MarkRedirectContextInUse(PTR_CONTEXT pCtx) { LIMITED_METHOD_CONTRACT; #ifdef _DEBUG _ASSERTE(!m_RedirectContextInUse); _ASSERTE(pCtx == m_pSavedRedirectContext); m_RedirectContextInUse = true; #endif } bool UnmarkRedirectContextInUse(PTR_CONTEXT pCtx) { LIMITED_METHOD_CONTRACT; #ifdef _DEBUG _ASSERTE(!UseContextBasedThreadRedirection() || m_RedirectContextInUse); if (m_RedirectContextInUse) { _ASSERTE(pCtx == m_pSavedRedirectContext); m_RedirectContextInUse = false; } #endif return (pCtx == m_pSavedRedirectContext); } #endif //DACCESS_COMPILE PTR_ThreadLocalData m_ThreadLocalDataPtr; int32_t cLoaderHandles = 0; PTR_LOADERHANDLE pLoaderHandles = 0; SpinLock m_TlsSpinLock; PTR_ThreadLocalData GetThreadLocalDataPtr() { LIMITED_METHOD_DAC_CONTRACT; return m_ThreadLocalDataPtr; } private: TailCallTls m_tailCallTls; public: TailCallTls* GetTailCallTls() { return &m_tailCallTls; } void* GetReturnAddress(void** retAddrSlot) { #ifdef FEATURE_HIJACK if ((m_State & TS_Hijacked) && (retAddrSlot == m_ppvHJRetAddrPtr)) { return m_pvHJRetAddr; } #endif return *retAddrSlot; } #ifdef FEATURE_HIJACK void* GetHijackedReturnAddress() { return m_pvHJRetAddr; } #endif #ifdef _DEBUG private: // When we create an object, or create an OBJECTREF, or create an Interior Pointer, or enter EE from managed // code, we will set this flag. // Inside GCHeapUtilities::StressHeap, we only do GC if this flag is TRUE. Then we reset it to zero. BOOL m_fStressHeapCount; public: void EnableStressHeap() { LIMITED_METHOD_CONTRACT; m_fStressHeapCount = TRUE; } void DisableStressHeap() { LIMITED_METHOD_CONTRACT; m_fStressHeapCount = FALSE; } BOOL StressHeapIsEnabled() { LIMITED_METHOD_CONTRACT; return m_fStressHeapCount; } size_t *m_pCleanedStackBase; #endif #ifdef DACCESS_COMPILE public: void EnumMemoryRegions(CLRDataEnumMemoryFlags flags); void EnumMemoryRegionsWorker(CLRDataEnumMemoryFlags flags); #endif public: // Is the current thread currently executing within a constrained execution region? static BOOL IsExecutingWithinCer(); #ifdef _DEBUG // when the thread is doing a stressing GC, some Crst violation could be ignored, by a non-elegant solution. private: BOOL m_bGCStressing; // the flag to indicate if the thread is doing a stressing GC BOOL m_bUniqueStacking; // the flag to indicate if the thread is doing a UniqueStack public: BOOL GetGCStressing () { return m_bGCStressing; } BOOL GetUniqueStacking () { return m_bUniqueStacking; } #endif private: //----------------------------------------------------------------------------- // AVInRuntimeImplOkay : its okay to have an AV in Runtime implementation while // this holder is in effect. // // { // AVInRuntimeImplOkayHolder foo(); // } // make AV's in the Runtime illegal on out of scope. //----------------------------------------------------------------------------- DWORD m_dwAVInRuntimeImplOkayCount; static void AVInRuntimeImplOkayAcquire(Thread * pThread) { LIMITED_METHOD_CONTRACT; if (pThread) { _ASSERTE(pThread->m_dwAVInRuntimeImplOkayCount != (DWORD)-1); pThread->m_dwAVInRuntimeImplOkayCount++; } } static void AVInRuntimeImplOkayRelease(Thread * pThread) { LIMITED_METHOD_CONTRACT; if (pThread) { _ASSERTE(pThread->m_dwAVInRuntimeImplOkayCount > 0); pThread->m_dwAVInRuntimeImplOkayCount--; } } public: static BOOL AVInRuntimeImplOkay(void) { LIMITED_METHOD_CONTRACT; Thread * pThread = GetThreadNULLOk(); if (pThread) { return (pThread->m_dwAVInRuntimeImplOkayCount > 0); } else { return FALSE; } } class AVInRuntimeImplOkayHolder { Thread * const m_pThread; public: AVInRuntimeImplOkayHolder() : m_pThread(GetThreadNULLOk()) { LIMITED_METHOD_CONTRACT; AVInRuntimeImplOkayAcquire(m_pThread); } AVInRuntimeImplOkayHolder(Thread * pThread) : m_pThread(pThread) { LIMITED_METHOD_CONTRACT; AVInRuntimeImplOkayAcquire(m_pThread); } ~AVInRuntimeImplOkayHolder() { LIMITED_METHOD_CONTRACT; AVInRuntimeImplOkayRelease(m_pThread); } }; #ifdef _DEBUG private: DWORD m_dwLockCount; public: void IncLockCount() { LIMITED_METHOD_CONTRACT; _ASSERTE (m_dwLockCount != (DWORD)-1); m_dwLockCount ++; } void DecLockCount() { LIMITED_METHOD_CONTRACT; _ASSERTE (m_dwLockCount > 0); m_dwLockCount --; } BOOL HasLock() const { LIMITED_METHOD_CONTRACT; return m_dwLockCount != 0; } DWORD GetLockCount() const { LIMITED_METHOD_CONTRACT; return m_dwLockCount; } #endif // _DEBUG private: // If HasStarted fails, we cache the exception here, and rethrow on the thread which // calls Thread.Start. Exception* m_pExceptionDuringStartup; public: OBJECTREF GetExceptionDuringStartup(); #ifdef HAVE_GCCOVER private: BYTE* m_pbDestCode; BYTE* m_pbSrcCode; #if defined(GCCOVER_TOLERATE_SPURIOUS_AV) LPVOID m_pLastAVAddress; #endif // defined(GCCOVER_TOLERATE_SPURIOUS_AV) public: void CommitGCStressInstructionUpdate(); void PostGCStressInstructionUpdate(BYTE* pbDestCode, BYTE* pbSrcCode) { LIMITED_METHOD_CONTRACT; PRECONDITION(!HasPendingGCStressInstructionUpdate()); VolatileStoreWithoutBarrier(&m_pbSrcCode, pbSrcCode); VolatileStore(&m_pbDestCode, pbDestCode); } bool HasPendingGCStressInstructionUpdate() { LIMITED_METHOD_CONTRACT; BYTE* dest = VolatileLoad(&m_pbDestCode); return dest != NULL; } bool TryClearGCStressInstructionUpdate(BYTE** ppbDestCode, BYTE** ppbSrcCode) { LIMITED_METHOD_CONTRACT; bool result = false; if(HasPendingGCStressInstructionUpdate()) { *ppbDestCode = InterlockedExchangeT(&m_pbDestCode, NULL); if(*ppbDestCode != NULL) { result = true; *ppbSrcCode = InterlockedExchangeT(&m_pbSrcCode, NULL); CONSISTENCY_CHECK(*ppbSrcCode != NULL); } } return result; } #if defined(GCCOVER_TOLERATE_SPURIOUS_AV) void SetLastAVAddress(LPVOID address) { LIMITED_METHOD_CONTRACT; m_pLastAVAddress = address; } LPVOID GetLastAVAddress() { LIMITED_METHOD_CONTRACT; return m_pLastAVAddress; } #endif // defined(GCCOVER_TOLERATE_SPURIOUS_AV) #endif // HAVE_GCCOVER private: // Although this is a pointer, it is used as a flag to indicate the current context is unsafe // to inspect. When NULL the context is safe to use, otherwise it points to the active patch skipper // and the context is unsafe to use. When running a patch skipper we could be in one of two // debug-only situations that the context inspecting/modifying code isn't generally prepared // to deal with. // a) We have set the IP to point somewhere in the patch skip table but have not yet run the // instruction // b) We executed the instruction in the patch skip table and now the IP could be anywhere // The debugger may need to fix up the IP to compensate for the instruction being run // from a different address. VolatilePtr m_debuggerActivePatchSkipper; public: VOID BeginDebuggerPatchSkip(DebuggerPatchSkip* patchSkipper) { LIMITED_METHOD_CONTRACT; _ASSERTE(!m_debuggerActivePatchSkipper.Load()); InterlockedExchangeT(m_debuggerActivePatchSkipper.GetPointer(), patchSkipper); _ASSERTE(m_debuggerActivePatchSkipper.Load()); } VOID EndDebuggerPatchSkip() { LIMITED_METHOD_CONTRACT; _ASSERTE(m_debuggerActivePatchSkipper.Load()); InterlockedExchangeT(m_debuggerActivePatchSkipper.GetPointer(), NULL); _ASSERTE(!m_debuggerActivePatchSkipper.Load()); } bool HasActivePatchSkip() const { LIMITED_METHOD_DAC_CONTRACT; return m_debuggerActivePatchSkipper.Load() != NULL; } private: static BOOL EnterWorkingOnThreadContext(Thread *pThread) { LIMITED_METHOD_CONTRACT; if(pThread->m_debuggerActivePatchSkipper.Load() != NULL) { return FALSE; } return TRUE; } static void LeaveWorkingOnThreadContext(Thread *pThread) { LIMITED_METHOD_CONTRACT; } typedef ConditionalStateHolder WorkingOnThreadContextHolder; private: BOOL m_fAllowProfilerCallbacks; public: // // These two methods are for profiler support. The profiler clears the allowed // value once it has delivered a ThreadDestroyed callback, so that it does not // deliver any notifications to the profiler afterwards which reference this // thread. Callbacks on this thread which do not reference this thread are // allowable. // BOOL ProfilerCallbacksAllowed(void) { return m_fAllowProfilerCallbacks; } void SetProfilerCallbacksAllowed(BOOL fValue) { m_fAllowProfilerCallbacks = fValue; } private: // Inside a host, we don't own a thread handle, and we avoid DuplicateHandle call. // If a thread is dying after we obtain the thread handle, our SuspendThread may fail // because the handle may be closed and reused for a completely different type of handle. // To solve this problem, we have a counter m_dwThreadHandleBeingUsed. Before we grab // the thread handle, we increment the counter. Before we return a thread back to SQL // in Reset and ExitTask, we wait until the counter drops to 0. Volatile m_dwThreadHandleBeingUsed; private: static BOOL s_fCleanFinalizedThread; public: #ifndef DACCESS_COMPILE static void SetCleanupNeededForFinalizedThread() { LIMITED_METHOD_CONTRACT; _ASSERTE (IsFinalizerThread()); s_fCleanFinalizedThread = TRUE; } #endif //!DACCESS_COMPILE static BOOL CleanupNeededForFinalizedThread() { LIMITED_METHOD_CONTRACT; return s_fCleanFinalizedThread; } private: // When we create throwable for an exception, we need to run managed code. // If the same type of exception is thrown while creating managed object, like InvalidProgramException, // we may be in an infinite recursive case. Exception *m_pCreatingThrowableForException; friend OBJECTREF CLRException::GetThrowable(); private: // At the end of a catch, we may raise ThreadAbortException. If catch clause set IP to resume in the // corresponding try block, our exception system will execute the same catch clause again and again. // So we save reference to the clause post which TA was reraised, which is used in ExInfo::ProcessManagedCallFrame // to make ThreadAbort proceed ahead instead of going in a loop. // This problem only happens on Win64 due to JIT64. The common scenario is VB's "On error resume next" DWORD m_dwIndexClauseForCatch; StackFrame m_sfEstablisherOfActualHandlerFrame; private: // Disables pumping and thread join in RCW creation bool m_fDisableComObjectEagerCleanup; // See ThreadStore::TriggerGCForDeadThreadsIfNecessary() bool m_fHasDeadThreadBeenConsideredForGCTrigger; #ifdef FEATURE_COMINTEROP private: // Cookie returned from CoRegisterInitializeSpy ULARGE_INTEGER m_uliInitializeSpyCookie; // True if m_uliInitializeSpyCookie is valid bool m_fInitializeSpyRegistered; // The last STA COM context we saw - used to speed up RCW creation LPVOID m_pLastSTACtxCookie; public: inline void RevokeApartmentSpy(); inline LPVOID GetLastSTACtxCookie(BOOL *pfNAContext); inline void SetLastSTACtxCookie(LPVOID pCtxCookie, BOOL fNAContext); #endif // FEATURE_COMINTEROP private: // This duplicates the ThreadType_GC bit stored in TLS (t_ThreadType). It exists // so that any thread can query whether any other thread is a "GC Special" thread. // (In contrast, ::IsGCSpecialThread() only gives this info about the currently // executing thread.) The Profiling API uses this to determine whether it should // "hide" the thread from profilers. GC Special threads (in particular the bgc // thread) need to be hidden from profilers because the bgc thread creation path // occurs while the EE is suspended, and while the thread that's suspending the // runtime is waiting for the bgc thread to signal an event. The bgc thread cannot // switch to preemptive mode and call into a profiler at this time, or else a // deadlock will result when toggling back to cooperative mode (bgc thread toggling // to coop will block due to the suspension, and the thread suspending the runtime // continues to block waiting for the bgc thread to signal its creation events). // Furthermore, profilers have no need to be aware of GC special threads anyway, // since managed code never runs on them. bool m_fGCSpecial; public: // Profiling API uses this to determine whether it should hide this thread from the // profiler. bool IsGCSpecial(); // GC calls this when creating special threads that also happen to have an EE Thread // object associated with them (e.g., the bgc thread). void SetGCSpecial(); private: PTR_GCFrame m_pGCFrame; // The topmost GC Frame #ifndef TARGET_UNIX WORD m_wCPUGroup; DWORD_PTR m_pAffinityMask; #endif // !TARGET_UNIX public: void ChooseThreadCPUGroupAffinity(); void ClearThreadCPUGroupAffinity(); private: // Per thread table used to implement allocation sampling. AllLoggedTypes * m_pAllLoggedTypes; public: AllLoggedTypes * GetAllocationSamplingTable() { LIMITED_METHOD_CONTRACT; return m_pAllLoggedTypes; } void SetAllocationSamplingTable(AllLoggedTypes * pAllLoggedTypes) { LIMITED_METHOD_CONTRACT; // Assert if we try to set the m_pAllLoggedTypes to a non NULL value if it is already non-NULL. // This implies a memory leak. _ASSERTE(pAllLoggedTypes != NULL ? m_pAllLoggedTypes == NULL : TRUE); m_pAllLoggedTypes = pAllLoggedTypes; } #ifdef FEATURE_PERFTRACING private: // The activity ID for the current thread. // An activity ID of zero means the thread is not executing in the context of an activity. GUID m_activityId; public: LPCGUID GetActivityId() const { LIMITED_METHOD_CONTRACT; return &m_activityId; } void SetActivityId(LPCGUID pActivityId) { LIMITED_METHOD_CONTRACT; _ASSERTE(pActivityId != NULL); m_activityId = *pActivityId; } #endif // FEATURE_PERFTRACING #ifdef FEATURE_HIJACK #ifdef TARGET_X86 private: // By the time a frame is scanned by the runtime, m_pHijackReturnKind always // identifies the gc-ness of the return register(s) ReturnKind m_HijackReturnKind; bool m_HijackHasAsyncRet; public: ReturnKind GetHijackReturnKind(bool* hasAsyncRet) { LIMITED_METHOD_CONTRACT; *hasAsyncRet = m_HijackHasAsyncRet; return m_HijackReturnKind; } void SetHijackReturnKind(ReturnKind returnKind, bool hasAsyncRet) { LIMITED_METHOD_CONTRACT; m_HijackReturnKind = returnKind; m_HijackHasAsyncRet = hasAsyncRet; } #endif #endif // FEATURE_HIJACK public: static uint64_t dead_threads_non_alloc_bytes; #ifndef DACCESS_COMPILE public: class CurrentPrepareCodeConfigHolder { private: Thread *const m_thread; #ifdef _DEBUG PrepareCodeConfig *const m_config; #endif public: CurrentPrepareCodeConfigHolder(Thread *thread, PrepareCodeConfig *config); ~CurrentPrepareCodeConfigHolder(); }; public: PrepareCodeConfig *GetCurrentPrepareCodeConfig() const { LIMITED_METHOD_CONTRACT; return m_currentPrepareCodeConfig; } #endif // !DACCESS_COMPILE private: PrepareCodeConfig *m_currentPrepareCodeConfig; #ifndef DACCESS_COMPILE public: bool IsInForbidSuspendForDebuggerRegion() const { LIMITED_METHOD_CONTRACT; return m_isInForbidSuspendForDebuggerRegion; } void EnterForbidSuspendForDebuggerRegion(); void ExitForbidSuspendForDebuggerRegion(); #endif private: bool m_isInForbidSuspendForDebuggerRegion; #ifndef DACCESS_COMPILE public: static void StaticInitialize(); #if defined(TARGET_WINDOWS) static bool AreShadowStacksEnabled() { LIMITED_METHOD_CONTRACT; #if defined(TARGET_AMD64) // The SSP is null when CET shadow stacks are not enabled. On processors that don't support shadow stacks, this is a // no-op and the intrinsic returns 0. CET shadow stacks are enabled or disabled for all threads, so the result is the // same from any thread. return _rdsspq() != 0; #else // When implementing AreShadowStacksEnabled() on other architectures, review all the places where this is used. return false; #endif } #endif #ifdef FEATURE_SPECIAL_USER_MODE_APC private: static void InitializeSpecialUserModeApc(); static void ApcActivationCallback(ULONG_PTR Parameter); #endif public: static bool UseSpecialUserModeApc() { LIMITED_METHOD_CONTRACT; #ifdef FEATURE_SPECIAL_USER_MODE_APC return s_pfnQueueUserAPC2Proc != nullptr; #else return false; #endif } static bool UseContextBasedThreadRedirection() { LIMITED_METHOD_CONTRACT; #ifndef DISABLE_THREADSUSPEND return !UseSpecialUserModeApc(); #else return false; #endif } static bool UseRedirectForGcStress() { LIMITED_METHOD_CONTRACT; #ifdef USE_REDIRECT_FOR_GCSTRESS return UseContextBasedThreadRedirection(); #else return false; #endif } #ifdef FEATURE_SPECIAL_USER_MODE_APC private: static QueueUserAPC2Proc s_pfnQueueUserAPC2Proc; #endif #endif // !DACCESS_COMPILE private: bool m_hasPendingActivation; friend struct ::cdac_data; }; template<> struct cdac_data { static constexpr size_t Id = offsetof(Thread, m_ThreadId); static constexpr size_t OSId = offsetof(Thread, m_OSThreadId); static constexpr size_t State = offsetof(Thread, m_State); static constexpr size_t DebuggerControlledThreadState = offsetof(Thread, m_DebuggerControlledThreadState); static constexpr size_t PreemptiveGCDisabled = offsetof(Thread, m_fPreemptiveGCDisabled); static constexpr size_t RuntimeThreadLocals = offsetof(Thread, m_pRuntimeThreadLocals); static constexpr size_t Frame = offsetof(Thread, m_pFrame); static constexpr size_t GCFrame = offsetof(Thread, m_pGCFrame); static constexpr size_t CachedStackBase = offsetof(Thread, m_CacheStackBase); static constexpr size_t CachedStackLimit = offsetof(Thread, m_CacheStackLimit); static constexpr size_t ExposedObject = offsetof(Thread, m_ExposedObject); static constexpr size_t LastThrownObject = offsetof(Thread, m_LastThrownObjectHandle); static constexpr size_t LastThrownObjectIsUnhandled = offsetof(Thread, m_ltoIsUnhandled); static constexpr size_t Link = offsetof(Thread, m_pNext); static constexpr size_t ThreadLocalDataPtr = offsetof(Thread, m_ThreadLocalDataPtr); static constexpr size_t CurrentCustomDebuggerNotification = offsetof(Thread, m_hCurrNotification); static_assert(std::is_same().m_ExceptionState), ThreadExceptionState>::value, "Thread::m_ExceptionState is of type ThreadExceptionState"); static constexpr size_t ExceptionTracker = offsetof(Thread, m_ExceptionState) + offsetof(ThreadExceptionState, m_pCurrentTracker); static constexpr size_t DebuggerFilterContext = offsetof(Thread, m_debuggerFilterContext); static constexpr size_t InteropDebuggingHijacked = offsetof(Thread, m_fInteropDebuggingHijacked); #ifdef TARGET_WINDOWS static constexpr size_t ThreadHandle = offsetof(Thread, m_ThreadHandle); #endif #ifndef TARGET_UNIX static constexpr size_t UEWatsonBucketTrackerBuckets = offsetof(Thread, m_ExceptionState) + offsetof(ThreadExceptionState, m_UEWatsonBucketTracker) + offsetof(EHWatsonBucketTracker, m_WatsonUnhandledInfo.m_pUnhandledBuckets); #endif static_assert(State == 0, "Thread.NativeThread depends on Thread::m_State being the first field"); }; // End of class Thread typedef Thread::ForbidSuspendThreadHolder ForbidSuspendThreadHolder; typedef Thread::ThreadPreventAsyncHolder ThreadPreventAsyncHolder; typedef Thread::AVInRuntimeImplOkayHolder AVInRuntimeImplOkayHolder; BOOL RevertIfImpersonated(BOOL *bReverted, HANDLE *phToken); void UndoRevert(BOOL bReverted, HANDLE hToken); // --------------------------------------------------------------------------- // // The ThreadStore manages all the threads in the system. // // There is one ThreadStore in the system, available through // ThreadStore::m_pThreadStore. // --------------------------------------------------------------------------- typedef SListTail ThreadList; // The ThreadStore is a singleton class #define CHECK_ONE_STORE() _ASSERTE(this == ThreadStore::s_pThreadStore); typedef DPTR(class ThreadStore) PTR_ThreadStore; class ThreadStore { friend class Thread; friend class ThreadSuspend; friend class AppDomain; #ifdef DACCESS_COMPILE friend class ClrDataAccess; friend Thread* __stdcall DacGetThread(ULONG32 osThreadID); #endif public: ThreadStore(); static void InitThreadStore(); static void LockThreadStore(); static void UnlockThreadStore(); // Add a Thread to the ThreadStore static void AddThread(Thread *newThread); // RemoveThread finds the thread in the ThreadStore and discards it. static BOOL RemoveThread(Thread *target); // Transfer a thread from the unstarted to the started list. static void TransferStartedThread(Thread *target); // Before using the thread list, be sure to take the critical section. Otherwise // it can change underneath you, perhaps leading to an exception after Remove. // Prev==NULL to get the first entry in the list. static Thread *GetAllThreadList(Thread *Prev, ULONG mask, ULONG bits); static Thread *GetThreadList(Thread *Prev); // We shut down the EE when the last non-background thread terminates. This event // is used to signal the main thread when this condition occurs. void WaitForOtherThreads(); static void CheckForEEShutdown(); CLREvent m_TerminationEvent; // Have all the foreground threads completed? In other words, can we release // the main thread? BOOL OtherThreadsComplete() { LIMITED_METHOD_CONTRACT; _ASSERTE(m_ThreadCount - m_UnstartedThreadCount - m_DeadThreadCount - Thread::m_ActiveDetachCount + m_PendingThreadCount >= m_BackgroundThreadCount); return (m_ThreadCount - m_UnstartedThreadCount - m_DeadThreadCount - Thread::m_ActiveDetachCount + m_PendingThreadCount == m_BackgroundThreadCount); } // If you want to trap threads re-entering the EE (for debugging, // or Thread.Suspend() or whatever, you need to IncrementTrapReturningThreads(). When // you are finished snagging threads, call DecrementTrapReturningThreads(). This // counts internally. // // Of course, you must also fix RareDisablePreemptiveGC to do the right thing // when the trap occurs. static void IncrementTrapReturningThreads(); static void DecrementTrapReturningThreads(); static void SetThreadTrapForSuspension(); static void UnsetThreadTrapForSuspension(); static bool IsTrappingThreadsForSuspension(); private: // Enter and leave the critical section around the thread store. Clients should // use LockThreadStore and UnlockThreadStore. void Enter(); void Leave(); // Critical section for adding and removing threads to the store Crst m_Crst; // List of all the threads known to the ThreadStore (started & unstarted). ThreadList m_ThreadList; // m_ThreadCount is the count of all threads in m_ThreadList. This includes // background threads / unstarted threads / whatever. // // m_UnstartedThreadCount is the subset of m_ThreadCount that have not yet been // started. // // m_BackgroundThreadCount is the subset of m_ThreadCount that have been started // but which are running in the background. So this is a misnomer in the sense // that unstarted background threads are not reflected in this count. // // m_PendingThreadCount is used to solve a race condition. The main thread could // start another thread running and then exit. The main thread might then start // tearing down the EE before the new thread moves itself out of m_UnstartedThread- // Count in TransferStartedThread. This count is atomically bumped in // CreateNewThread, and atomically reduced within a locked thread store. // // m_DeadThreadCount is the subset of m_ThreadCount which have died. The Win32 // thread has disappeared, but something (like the exposed object) has kept the // refcount non-zero so we can't destruct yet. protected: LONG m_ThreadCount; public: LONG ThreadCountInEE () { LIMITED_METHOD_CONTRACT; return m_ThreadCount; } private: LONG m_UnstartedThreadCount; LONG m_BackgroundThreadCount; LONG m_PendingThreadCount; public: LONG GetPendingThreadCount () { LIMITED_METHOD_CONTRACT; return m_PendingThreadCount; } private: LONG m_DeadThreadCount; LONG m_DeadThreadCountForGCTrigger; bool m_TriggerGCForDeadThreads; private: // Even in the release product, we need to know what thread holds the lock on // the ThreadStore. This is so we never deadlock when the GC thread halts a // thread that holds this lock. Thread *m_HoldingThread; EEThreadId m_holderthreadid; // current holder (or NULL) private: static LONG s_DeadThreadCountThresholdForGCTrigger; static DWORD s_DeadThreadGCTriggerPeriodMilliseconds; static SIZE_T *s_DeadThreadGenerationCounts; public: static BOOL HoldingThreadStore() { WRAPPER_NO_CONTRACT; // Note that GetThread() may be 0 if it is the debugger thread // or perhaps a concurrent GC thread. return HoldingThreadStore(GetThreadNULLOk()); } static BOOL HoldingThreadStore(Thread *pThread); #ifdef DACCESS_COMPILE static void EnumMemoryRegions(CLRDataEnumMemoryFlags flags); #endif SPTR_DECL(ThreadStore, s_pThreadStore); #ifdef _DEBUG public: BOOL DbgFindThread(Thread *target); LONG DbgBackgroundThreadCount() { LIMITED_METHOD_CONTRACT; return m_BackgroundThreadCount; } BOOL IsCrstForThreadStore (const CrstBase* const pCrstBase) { LIMITED_METHOD_CONTRACT; return (void *)pCrstBase == (void*)&m_Crst; } #endif private: static BYTE* s_pOSContextBuffer; static CONTEXT *s_pOSContext; public: // Pre-allocate an OS context for possible use by a redirected thread and keep in a static variable. // // There are two reasons for this pattern: // - We can not do any memory allocation after we suspend a thread in order to avoid deadlock situation. // So, when anticipating a need, we must pre-allocate. // // - Even though we know the thread we are suspending, we do not want to put the context directly on the // thread because the thread only _may_ need the context. Often it does not end up needing it, // then we will keep the context for the next time like this. static void AllocateOSContext(); // Retrieves and detaches the pre-alocated context + optional containing buffer (when CONTEXT_XSTATE is used) static CONTEXT *GrabOSContext(BYTE** contextBuffer); private: // Thread abort needs to walk stack to decide if thread abort can proceed. // It is unsafe to crawl a stack of thread if the thread is OS-suspended which we do during // thread abort. For example, Thread T1 aborts thread T2. T2 is suspended by T1. Inside SQL // this means that no thread sharing the same scheduler with T2 can run. If T1 needs a lock which // is owned by one thread on the scheduler, T1 will wait forever. // Our solution is to move T2 to a safe point, resume it, and then do stack crawl. static CLREvent *s_pWaitForStackCrawlEvent; public: static void WaitForStackCrawlEvent() { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_ANY; CAN_TAKE_LOCK; } CONTRACTL_END; s_pWaitForStackCrawlEvent->Wait(INFINITE,FALSE); } static void SetStackCrawlEvent() { LIMITED_METHOD_CONTRACT; s_pWaitForStackCrawlEvent->Set(); } static void ResetStackCrawlEvent() { LIMITED_METHOD_CONTRACT; s_pWaitForStackCrawlEvent->Reset(); } private: void IncrementDeadThreadCountForGCTrigger(); void DecrementDeadThreadCountForGCTrigger(); public: void OnMaxGenerationGCStarted(); bool ShouldTriggerGCForDeadThreads(); void TriggerGCForDeadThreadsIfNecessary(); friend struct ::cdac_data; }; template<> struct cdac_data { static constexpr size_t FirstThreadLink = offsetof(ThreadStore, m_ThreadList) + offsetof(ThreadList, m_pHead); static constexpr size_t ThreadCount = offsetof(ThreadStore, m_ThreadCount); static constexpr size_t UnstartedCount = offsetof(ThreadStore, m_UnstartedThreadCount); static constexpr size_t BackgroundCount = offsetof(ThreadStore, m_BackgroundThreadCount); static constexpr size_t PendingCount = offsetof(ThreadStore, m_PendingThreadCount); static constexpr size_t DeadCount = offsetof(ThreadStore, m_DeadThreadCount); }; typedef StateHolder ThreadStoreLockHolder; // This class dispenses small thread ids for the thin lock mechanism. class IdDispenser { private: DWORD m_highestId; // highest id given out so far SIZE_T m_recycleBin; // link list to chain all ids returning to us Crst m_Crst; // lock to protect our data structures DPTR(PTR_Thread) m_idToThread; // map thread ids to threads DWORD m_idToThreadCapacity; // capacity of the map #ifndef DACCESS_COMPILE void GrowIdToThread() { CONTRACTL { THROWS; GC_NOTRIGGER; MODE_ANY; } CONTRACTL_END; DWORD newCapacity = m_idToThreadCapacity == 0 ? 16 : m_idToThreadCapacity*2; Thread **newIdToThread = new Thread*[newCapacity]; newIdToThread[0] = NULL; for (DWORD i = 1; i < m_idToThreadCapacity; i++) { newIdToThread[i] = m_idToThread[i]; } for (DWORD j = m_idToThreadCapacity; j < newCapacity; j++) { newIdToThread[j] = NULL; } delete[] m_idToThread; m_idToThread = newIdToThread; m_idToThreadCapacity = newCapacity; } #endif // !DACCESS_COMPILE public: IdDispenser() : // NOTE: CRST_UNSAFE_ANYMODE prevents a GC mode switch when entering this crst. // If you remove this flag, we will switch to preemptive mode when entering // m_Crst, which means all functions that enter it will become // GC_TRIGGERS. (This includes all uses of CrstHolder.) So be sure // to update the contracts if you remove this flag. m_Crst(CrstThreadIdDispenser, CRST_UNSAFE_ANYMODE) { WRAPPER_NO_CONTRACT; m_highestId = 0; m_recycleBin = 0; m_idToThreadCapacity = 0; m_idToThread = NULL; } ~IdDispenser() { LIMITED_METHOD_CONTRACT; delete[] m_idToThread; } bool IsValidId(DWORD id) { LIMITED_METHOD_CONTRACT; return (id > 0) && (id <= m_highestId); } #ifndef DACCESS_COMPILE void NewId(Thread *pThread, DWORD & newId) { WRAPPER_NO_CONTRACT; DWORD result; CrstHolder ch(&m_Crst); if (m_recycleBin != 0) { _ASSERTE(FitsIn(m_recycleBin)); result = static_cast(m_recycleBin); m_recycleBin = reinterpret_cast(m_idToThread[m_recycleBin]); } else { // we make sure ids don't wrap around - before they do, we always return the highest possible // one and rely on our caller to detect this situation if (m_highestId + 1 > m_highestId) m_highestId = m_highestId + 1; result = m_highestId; if (result >= m_idToThreadCapacity) GrowIdToThread(); } _ASSERTE(result < m_idToThreadCapacity); newId = result; if (result < m_idToThreadCapacity) m_idToThread[result] = pThread; } #endif // !DACCESS_COMPILE #ifndef DACCESS_COMPILE void DisposeId(DWORD id) { CONTRACTL { NOTHROW; GC_NOTRIGGER; MODE_ANY; CAN_TAKE_LOCK; } CONTRACTL_END; CrstHolder ch(&m_Crst); _ASSERTE(IsValidId(id)); if (id == m_highestId) { m_highestId--; } else { m_idToThread[id] = reinterpret_cast(m_recycleBin); m_recycleBin = id; #ifdef _DEBUG size_t index = (size_t)m_idToThread[id]; while (index != 0) { _ASSERTE(index != id); index = (size_t)m_idToThread[index]; } #endif } } #endif // !DACCESS_COMPILE Thread *IdToThread(DWORD id) { LIMITED_METHOD_CONTRACT; CrstHolder ch(&m_Crst); Thread *result = NULL; if (id <= m_highestId) result = m_idToThread[id]; // m_idToThread may have Thread*, or the next free slot _ASSERTE ((size_t)result > m_idToThreadCapacity); return result; } PTR_Thread IdToThreadWithValidation(DWORD id) { WRAPPER_NO_CONTRACT; CrstHolder ch(&m_Crst); PTR_Thread result = NULL; if (id <= m_highestId) result = m_idToThread[id]; // m_idToThread may have Thread*, or the next free slot if (dac_cast(result) <= m_idToThreadCapacity) result = NULL; _ASSERTE(result == NULL || (dac_cast(result) & 0x3) == 0 || ((Thread*)result)->GetThreadId() == id); return result; } friend struct ::cdac_data; }; template<> struct cdac_data { static constexpr size_t IdToThread = offsetof(IdDispenser, m_idToThread); static constexpr size_t HighestId = offsetof(IdDispenser, m_highestId); }; typedef DPTR(IdDispenser) PTR_IdDispenser; // Dispenser of small thread ids for thin lock mechanism GPTR_DECL(IdDispenser,g_pThinLockThreadIdDispenser); // forward declaration DWORD MsgWaitHelper(int numWaiters, HANDLE* phEvent, BOOL bWaitAll, DWORD millis, BOOL alertable = FALSE); // When a thread is being created after a debug suspension has started, it sends an event up to the // debugger. Afterwards, with the Debugger Lock still held, it will check to see if we had already asked to suspend the // Runtime. If we have, then it will turn around and call this to set the debug suspend pending flag on the newly // created thread, since it was missed by SysStartSuspendForDebug as it didn't exist when that function was run. // inline void Thread::MarkForDebugSuspend(void) { WRAPPER_NO_CONTRACT; if (!HasThreadState(TS_DebugSuspendPending)) { SetThreadState(TS_DebugSuspendPending); ThreadStore::IncrementTrapReturningThreads(); } } // Debugger per-thread flag for enabling notification on "manual" // method calls, for stepping logic. inline void Thread::IncrementTraceCallCount() { WRAPPER_NO_CONTRACT; InterlockedIncrement(&m_TraceCallCount); ThreadStore::IncrementTrapReturningThreads(); } inline void Thread::DecrementTraceCallCount() { WRAPPER_NO_CONTRACT; ThreadStore::DecrementTrapReturningThreads(); InterlockedDecrement(&m_TraceCallCount); } // -------------------------------------------------------------------------------- // GCHolder is used to implement the normal GCX_ macros. // // GCHolder is normally used indirectly through GCX_ convenience macros, but can be used // directly if needed (e.g. due to multiple holders in one scope, or to use // in class definitions). // // GCHolder (or derived types) should only be instantiated as automatic variables // -------------------------------------------------------------------------------- #ifdef ENABLE_CONTRACTS_IMPL #define GCHOLDER_CONTRACT_ARGS_NoDtor , false, szConstruct, szFunction, szFile, lineNum #define GCHOLDER_CONTRACT_ARGS_HasDtor , true, szConstruct, szFunction, szFile, lineNum #define GCHOLDER_DECLARE_CONTRACT_ARGS_BARE \ const char * szConstruct = "Unknown" \ , const char * szFunction = "Unknown" \ , const char * szFile = "Unknown" \ , int lineNum = 0 #define GCHOLDER_DECLARE_CONTRACT_ARGS , GCHOLDER_DECLARE_CONTRACT_ARGS_BARE #define GCHOLDER_DECLARE_CONTRACT_ARGS_INTERNAL , bool fPushStackRecord = true, GCHOLDER_DECLARE_CONTRACT_ARGS_BARE #define GCHOLDER_SETUP_CONTRACT_STACK_RECORD(mode) \ m_fPushedRecord = false; \ \ if (fPushStackRecord && conditional) \ { \ m_pClrDebugState = GetClrDebugState(); \ m_oldClrDebugState = *m_pClrDebugState; \ \ m_pClrDebugState->ViolationMaskReset( ModeViolation ); \ \ m_ContractStackRecord.m_szFunction = szFunction; \ m_ContractStackRecord.m_szFile = szFile; \ m_ContractStackRecord.m_lineNum = lineNum; \ m_ContractStackRecord.m_testmask = \ (Contract::ALL_Disabled & ~((UINT)(Contract::MODE_Mask))) \ | (mode); \ m_ContractStackRecord.m_construct = szConstruct; \ m_pClrDebugState->LinkContractStackTrace( &m_ContractStackRecord ); \ m_fPushedRecord = true; \ } #define GCHOLDER_CHECK_FOR_PREEMP_IN_NOTRIGGER(pThread) \ if (pThread->GCNoTrigger()) \ { \ CONTRACT_ASSERT("Coop->preemp->coop switch attempted in a GC_NOTRIGGER scope", \ Contract::GC_NoTrigger, \ Contract::GC_Mask, \ szFunction, \ szFile, \ lineNum \ ); \ } #else #define GCHOLDER_CONTRACT_ARGS_NoDtor #define GCHOLDER_CONTRACT_ARGS_HasDtor #define GCHOLDER_DECLARE_CONTRACT_ARGS_BARE #define GCHOLDER_DECLARE_CONTRACT_ARGS #define GCHOLDER_DECLARE_CONTRACT_ARGS_INTERNAL #define GCHOLDER_SETUP_CONTRACT_STACK_RECORD(mode) #define GCHOLDER_CHECK_FOR_PREEMP_IN_NOTRIGGER(pThread) #endif // ENABLE_CONTRACTS_IMPL #ifndef DACCESS_COMPILE class GCHolderBase { protected: // NOTE: This method is FORCEINLINE'ed into its callers, but the callers are just the // corresponding methods in the derived types, not all sites that use GC holders. This // is done so that the #pragma optimize will take affect since the optimize settings // are taken from the template instantiation site, not the template definition site. template FORCEINLINE_NONDEBUG void PopInternal() { WRAPPER_NO_CONTRACT; #ifdef ENABLE_CONTRACTS_IMPL if (m_fPushedRecord) { *m_pClrDebugState = m_oldClrDebugState; } // Make sure that we're using the version of this template that matches the // invariant setup in EnterInternal{Coop|Preemp}{_HackNoThread} _ASSERTE(!!THREAD_EXISTS == m_fThreadMustExist); #endif if (m_WasCoop) { // m_WasCoop is only TRUE if we've already verified there's an EE thread. _ASSERTE(m_Thread != NULL); // Cannot switch to cooperative with no thread if (!m_Thread->PreemptiveGCDisabled()) m_Thread->DisablePreemptiveGC(); } else { // Either we initialized m_Thread explicitly with GetThread() in the // constructor, or our caller (instantiator of GCHolder) called our constructor // with GetThread() (which we already asserted in the constructor) // (i.e., m_Thread == GetThread()). Also, note that if THREAD_EXISTS, // then m_Thread must be non-null (as it's == GetThread()). So the // "if" below looks a little hokey since we're checking for either condition. // But the template param THREAD_EXISTS allows us to statically early-out // when it's TRUE, so we check it for perf. if (THREAD_EXISTS || m_Thread != NULL) { if (m_Thread->PreemptiveGCDisabled()) m_Thread->EnablePreemptiveGC(); } } // If we have a thread then we assert that we ended up in the same state // which we started in. if (THREAD_EXISTS || m_Thread != NULL) { _ASSERTE(!!m_WasCoop == !!(m_Thread->PreemptiveGCDisabled())); } } // NOTE: The rest of these methods are all FORCEINLINE so that the uses where 'conditional==true' // can have the if-checks removed by the compiler. The callers are just the corresponding methods // in the derived types, not all sites that use GC holders. // This is broken - there is a potential race with the GC thread. It is currently // used for a few cases where (a) we potentially haven't started up the EE yet, or // (b) we are on a "special thread". We need a real solution here though. FORCEINLINE_NONDEBUG void EnterInternalCoop_HackNoThread(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS_INTERNAL) { GCHOLDER_SETUP_CONTRACT_STACK_RECORD(Contract::MODE_Coop); m_Thread = GetThreadNULLOk(); #ifdef ENABLE_CONTRACTS_IMPL m_fThreadMustExist = false; #endif // ENABLE_CONTRACTS_IMPL if (m_Thread != NULL) { m_WasCoop = m_Thread->PreemptiveGCDisabled(); if (conditional && !m_WasCoop) { m_Thread->DisablePreemptiveGC(); _ASSERTE(m_Thread->PreemptiveGCDisabled()); } } else { m_WasCoop = FALSE; } } FORCEINLINE_NONDEBUG void EnterInternalPreemp(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS_INTERNAL) { GCHOLDER_SETUP_CONTRACT_STACK_RECORD(Contract::MODE_Preempt); m_Thread = GetThreadNULLOk(); #ifdef ENABLE_CONTRACTS_IMPL m_fThreadMustExist = false; if (m_Thread != NULL && conditional) { GCHOLDER_CHECK_FOR_PREEMP_IN_NOTRIGGER(m_Thread); } #endif // ENABLE_CONTRACTS_IMPL if (m_Thread != NULL) { m_WasCoop = m_Thread->PreemptiveGCDisabled(); if (conditional && m_WasCoop) { m_Thread->EnablePreemptiveGC(); _ASSERTE(!m_Thread->PreemptiveGCDisabled()); } } else { m_WasCoop = FALSE; } } FORCEINLINE_NONDEBUG void EnterInternalCoop(Thread *pThread, bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS_INTERNAL) { // This is the perf version. So we deliberately restrict the calls // to already setup threads to avoid the null checks and GetThread call _ASSERTE(pThread == GetThread()); #ifdef ENABLE_CONTRACTS_IMPL m_fThreadMustExist = true; #endif // ENABLE_CONTRACTS_IMPL GCHOLDER_SETUP_CONTRACT_STACK_RECORD(Contract::MODE_Coop); m_Thread = pThread; m_WasCoop = m_Thread->PreemptiveGCDisabled(); if (conditional && !m_WasCoop) { m_Thread->DisablePreemptiveGC(); _ASSERTE(m_Thread->PreemptiveGCDisabled()); } } template FORCEINLINE_NONDEBUG void EnterInternalPreemp(Thread *pThread, bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS_INTERNAL) { // This is the perf version. So we deliberately restrict the calls // to already setup threads to avoid the null checks and GetThread call _ASSERTE(!THREAD_EXISTS || (pThread == GetThread())); #ifdef ENABLE_CONTRACTS_IMPL m_fThreadMustExist = !!THREAD_EXISTS; #endif // ENABLE_CONTRACTS_IMPL GCHOLDER_SETUP_CONTRACT_STACK_RECORD(Contract::MODE_Preempt); m_Thread = pThread; if (THREAD_EXISTS || (m_Thread != NULL)) { GCHOLDER_CHECK_FOR_PREEMP_IN_NOTRIGGER(m_Thread); m_WasCoop = m_Thread->PreemptiveGCDisabled(); if (conditional && m_WasCoop) { m_Thread->EnablePreemptiveGC(); _ASSERTE(!m_Thread->PreemptiveGCDisabled()); } } else { m_WasCoop = FALSE; } } private: Thread * m_Thread; BOOL m_WasCoop; // This is BOOL and not 'bool' because PreemptiveGCDisabled returns BOOL, // so the codegen is better if we don't have to convert to 'bool'. #ifdef ENABLE_CONTRACTS_IMPL bool m_fThreadMustExist; // used to validate that the proper Pop method is used bool m_fPushedRecord; ClrDebugState m_oldClrDebugState; ClrDebugState *m_pClrDebugState; ContractStackRecord m_ContractStackRecord; #endif }; class GCCoopNoDtor : public GCHolderBase { public: DEBUG_NOINLINE void Enter(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { WRAPPER_NO_CONTRACT; if (conditional) { STATIC_CONTRACT_MODE_COOPERATIVE; } // The thread must be non-null to enter MODE_COOP this->EnterInternalCoop(GetThread(), conditional GCHOLDER_CONTRACT_ARGS_NoDtor); } DEBUG_NOINLINE void Leave() { WRAPPER_NO_CONTRACT; this->PopInternal(); // Thread must be non-NULL } }; class GCPreempNoDtor : public GCHolderBase { public: DEBUG_NOINLINE void Enter(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_PREEMPTIVE; } this->EnterInternalPreemp(conditional GCHOLDER_CONTRACT_ARGS_NoDtor); } DEBUG_NOINLINE void Enter(Thread * pThreadNullOk, bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_PREEMPTIVE; } this->EnterInternalPreemp( // Thread may be NULL pThreadNullOk, conditional GCHOLDER_CONTRACT_ARGS_NoDtor); } DEBUG_NOINLINE void Leave() { this->PopInternal(); // Thread may be NULL } }; class GCCoop : public GCHolderBase { public: DEBUG_NOINLINE GCCoop(GCHOLDER_DECLARE_CONTRACT_ARGS_BARE) { STATIC_CONTRACT_MODE_COOPERATIVE; // The thread must be non-null to enter MODE_COOP this->EnterInternalCoop(GetThread(), true GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE GCCoop(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_COOPERATIVE; } // The thread must be non-null to enter MODE_COOP this->EnterInternalCoop(GetThread(), conditional GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE ~GCCoop() { this->PopInternal(); // Thread must be non-NULL } }; // This is broken - there is a potential race with the GC thread. It is currently // used for a few cases where (a) we potentially haven't started up the EE yet, or // (b) we are on a "special thread". We need a real solution here though. class GCCoopHackNoThread : public GCHolderBase { public: DEBUG_NOINLINE GCCoopHackNoThread(GCHOLDER_DECLARE_CONTRACT_ARGS_BARE) { STATIC_CONTRACT_MODE_COOPERATIVE; this->EnterInternalCoop_HackNoThread(true GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE GCCoopHackNoThread(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_COOPERATIVE; } this->EnterInternalCoop_HackNoThread(conditional GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE ~GCCoopHackNoThread() { this->PopInternal(); // Thread might be NULL } }; class GCCoopThreadExists : public GCHolderBase { public: DEBUG_NOINLINE GCCoopThreadExists(Thread * pThread GCHOLDER_DECLARE_CONTRACT_ARGS) { STATIC_CONTRACT_MODE_COOPERATIVE; this->EnterInternalCoop(pThread, true GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE GCCoopThreadExists(Thread * pThread, bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_COOPERATIVE; } this->EnterInternalCoop(pThread, conditional GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE ~GCCoopThreadExists() { this->PopInternal(); // Thread must be non-NULL } }; class GCPreemp : public GCHolderBase { public: DEBUG_NOINLINE GCPreemp(GCHOLDER_DECLARE_CONTRACT_ARGS_BARE) { STATIC_CONTRACT_MODE_PREEMPTIVE; this->EnterInternalPreemp(true GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE GCPreemp(bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_PREEMPTIVE; } this->EnterInternalPreemp(conditional GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE ~GCPreemp() { this->PopInternal(); // Thread may be NULL } }; class GCPreempThreadExists : public GCHolderBase { public: DEBUG_NOINLINE GCPreempThreadExists(Thread * pThread GCHOLDER_DECLARE_CONTRACT_ARGS) { STATIC_CONTRACT_MODE_PREEMPTIVE; this->EnterInternalPreemp( // Thread must be non-NULL pThread, true GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE GCPreempThreadExists(Thread * pThread, bool conditional GCHOLDER_DECLARE_CONTRACT_ARGS) { if (conditional) { STATIC_CONTRACT_MODE_PREEMPTIVE; } this->EnterInternalPreemp( // Thread must be non-NULL pThread, conditional GCHOLDER_CONTRACT_ARGS_HasDtor); } DEBUG_NOINLINE ~GCPreempThreadExists() { this->PopInternal(); // Thread must be non-NULL } }; #endif // DACCESS_COMPILE // -------------------------------------------------------------------------------- // GCAssert is used to implement the assert GCX_ macros. Usage is similar to GCHolder. // // GCAsserting for preemptive mode automatically passes on unmanaged threads. // // Note that the assert is "2 sided"; it happens on entering and on leaving scope, to // help ensure mode integrity. // // GCAssert is a noop in a free build // -------------------------------------------------------------------------------- template class GCAssert { public: DEBUG_NOINLINE void BeginGCAssert(); DEBUG_NOINLINE void EndGCAssert() { } }; template class AutoCleanupGCAssert { #ifdef _DEBUG_IMPL public: DEBUG_NOINLINE AutoCleanupGCAssert(); DEBUG_NOINLINE ~AutoCleanupGCAssert() { WRAPPER_NO_CONTRACT; // This is currently disabled; we currently have a lot of code which doesn't // back out the GC mode properly (instead relying on the EX_TRY macros.) // // @todo enable this when we remove raw GC mode switching. #if 0 DoCheck(); #endif } private: FORCEINLINE void DoCheck() { WRAPPER_NO_CONTRACT; Thread *pThread = GetThreadNULLOk(); if (COOPERATIVE) { _ASSERTE(pThread != NULL); _ASSERTE(pThread->PreemptiveGCDisabled()); } else { _ASSERTE(pThread == NULL || !(pThread->PreemptiveGCDisabled())); } } #endif }; // -------------------------------------------------------------------------------- // GCForbid is used to add ForbidGC semantics to the current GC mode. Note that // it requires the thread to be in cooperative mode already. // // GCForbid is a noop in a free build // -------------------------------------------------------------------------------- #ifndef DACCESS_COMPILE class GCForbid : AutoCleanupGCAssert { #ifdef ENABLE_CONTRACTS_IMPL public: DEBUG_NOINLINE GCForbid(BOOL fConditional, const char *szFunction, const char *szFile, int lineNum) { if (fConditional) { STATIC_CONTRACT_MODE_COOPERATIVE; STATIC_CONTRACT_GC_NOTRIGGER; } m_fConditional = fConditional; if (m_fConditional) { m_pClrDebugState = ::GetClrDebugState(); m_oldClrDebugState = *m_pClrDebugState; m_pClrDebugState->ViolationMaskReset( GCViolation ); GetThread()->BeginForbidGC(szFile, lineNum); m_ContractStackRecord.m_szFunction = szFunction; m_ContractStackRecord.m_szFile = (char*)szFile; m_ContractStackRecord.m_lineNum = lineNum; m_ContractStackRecord.m_testmask = (Contract::ALL_Disabled & ~((UINT)(Contract::GC_Mask))) | Contract::GC_NoTrigger; m_ContractStackRecord.m_construct = "GCX_FORBID"; m_pClrDebugState->LinkContractStackTrace( &m_ContractStackRecord ); } } DEBUG_NOINLINE GCForbid(const char *szFunction, const char *szFile, int lineNum) { STATIC_CONTRACT_MODE_COOPERATIVE; STATIC_CONTRACT_GC_NOTRIGGER; m_fConditional = TRUE; m_pClrDebugState = ::GetClrDebugState(); m_oldClrDebugState = *m_pClrDebugState; m_pClrDebugState->ViolationMaskReset( GCViolation ); GetThread()->BeginForbidGC(szFile, lineNum); m_ContractStackRecord.m_szFunction = szFunction; m_ContractStackRecord.m_szFile = (char*)szFile; m_ContractStackRecord.m_lineNum = lineNum; m_ContractStackRecord.m_testmask = (Contract::ALL_Disabled & ~((UINT)(Contract::GC_Mask))) | Contract::GC_NoTrigger; m_ContractStackRecord.m_construct = "GCX_FORBID"; m_pClrDebugState->LinkContractStackTrace( &m_ContractStackRecord ); } DEBUG_NOINLINE ~GCForbid() { if (m_fConditional) { GetThread()->EndForbidGC(); *m_pClrDebugState = m_oldClrDebugState; } } private: BOOL m_fConditional; ClrDebugState *m_pClrDebugState; ClrDebugState m_oldClrDebugState; ContractStackRecord m_ContractStackRecord; #endif // _DEBUG_IMPL }; #endif // !DACCESS_COMPILE // -------------------------------------------------------------------------------- // GCNoTrigger is used to add NoTriggerGC semantics to the current GC mode. Unlike // GCForbid, it does not require a thread to be in cooperative mode. // // GCNoTrigger is a noop in a free build // -------------------------------------------------------------------------------- #ifndef DACCESS_COMPILE class GCNoTrigger { #ifdef ENABLE_CONTRACTS_IMPL public: DEBUG_NOINLINE GCNoTrigger(BOOL fConditional, const char *szFunction, const char *szFile, int lineNum) { if (fConditional) { STATIC_CONTRACT_GC_NOTRIGGER; } m_fConditional = fConditional; if (m_fConditional) { Thread * pThread = GetThreadNULLOk(); m_pClrDebugState = ::GetClrDebugState(); m_oldClrDebugState = *m_pClrDebugState; m_pClrDebugState->ViolationMaskReset( GCViolation ); if (pThread != NULL) { pThread->BeginNoTriggerGC(szFile, lineNum); } m_ContractStackRecord.m_szFunction = szFunction; m_ContractStackRecord.m_szFile = (char*)szFile; m_ContractStackRecord.m_lineNum = lineNum; m_ContractStackRecord.m_testmask = (Contract::ALL_Disabled & ~((UINT)(Contract::GC_Mask))) | Contract::GC_NoTrigger; m_ContractStackRecord.m_construct = "GCX_NOTRIGGER"; m_pClrDebugState->LinkContractStackTrace( &m_ContractStackRecord ); } } DEBUG_NOINLINE GCNoTrigger(const char *szFunction, const char *szFile, int lineNum) { STATIC_CONTRACT_GC_NOTRIGGER; m_fConditional = TRUE; Thread * pThread = GetThreadNULLOk(); m_pClrDebugState = ::GetClrDebugState(); m_oldClrDebugState = *m_pClrDebugState; m_pClrDebugState->ViolationMaskReset( GCViolation ); if (pThread != NULL) { pThread->BeginNoTriggerGC(szFile, lineNum); } m_ContractStackRecord.m_szFunction = szFunction; m_ContractStackRecord.m_szFile = (char*)szFile; m_ContractStackRecord.m_lineNum = lineNum; m_ContractStackRecord.m_testmask = (Contract::ALL_Disabled & ~((UINT)(Contract::GC_Mask))) | Contract::GC_NoTrigger; m_ContractStackRecord.m_construct = "GCX_NOTRIGGER"; m_pClrDebugState->LinkContractStackTrace( &m_ContractStackRecord ); } DEBUG_NOINLINE ~GCNoTrigger() { if (m_fConditional) { Thread * pThread = GetThreadNULLOk(); if (pThread) { pThread->EndNoTriggerGC(); } *m_pClrDebugState = m_oldClrDebugState; } } private: BOOL m_fConditional; ClrDebugState *m_pClrDebugState; ClrDebugState m_oldClrDebugState; ContractStackRecord m_ContractStackRecord; #endif // _DEBUG_IMPL }; #endif //!DACCESS_COMPILE class CoopTransitionHolder { Frame * m_pFrame; #ifdef DEBUG int m_uncaughtExceptions; #endif public: CoopTransitionHolder(Thread * pThread) : m_pFrame(pThread->m_pFrame) { LIMITED_METHOD_CONTRACT; #ifdef DEBUG m_uncaughtExceptions = std::uncaught_exceptions(); #endif } ~CoopTransitionHolder() { WRAPPER_NO_CONTRACT; _ASSERTE_MSG(m_pFrame == nullptr || m_uncaughtExceptions < std::uncaught_exceptions(), "Early return from JIT/EE interface method"); if (m_pFrame != nullptr) COMPlusCooperativeTransitionHandler(m_pFrame); } void SuppressRelease() { LIMITED_METHOD_CONTRACT; // FRAME_TOP and NULL must be distinct values. // static_assert(FRAME_TOP_VALUE != NULL); m_pFrame = nullptr; } }; // -------------------------------------------------------------------------------- // GCX macros - see util.hpp // -------------------------------------------------------------------------------- #ifdef _DEBUG_IMPL // Normally, any thread we operate on has a Thread block in its TLS. But there are // a few special threads we don't normally execute managed code on. BOOL dbgOnly_IsSpecialEEThread(); void dbgOnly_IdentifySpecialEEThread(); #ifdef USE_CHECKED_OBJECTREFS #define ASSERT_PROTECTED(objRef) Thread::ObjectRefProtected(objRef) #else #define ASSERT_PROTECTED(objRef) #endif #else #define ASSERT_PROTECTED(objRef) #endif #ifdef ENABLE_CONTRACTS_IMPL #define BEGINFORBIDGC() {if (GetThreadNULLOk() != NULL) GetThreadNULLOk()->BeginForbidGC(__FILE__, __LINE__);} #define ENDFORBIDGC() {if (GetThreadNULLOk() != NULL) GetThreadNULLOk()->EndForbidGC();} class FCallGCCanTrigger { public: static DEBUG_NOINLINE void Enter() { STATIC_CONTRACT_GC_TRIGGERS; Thread * pThread = GetThreadNULLOk(); if (pThread != NULL) { Enter(pThread); } } static DEBUG_NOINLINE void Enter(Thread* pThread) { STATIC_CONTRACT_GC_TRIGGERS; pThread->EndForbidGC(); } static DEBUG_NOINLINE void Leave(const char *szFunction, const char *szFile, int lineNum) { Thread * pThread = GetThreadNULLOk(); if (pThread != NULL) { Leave(pThread, szFunction, szFile, lineNum); } } static DEBUG_NOINLINE void Leave(Thread* pThread, const char *szFunction, const char *szFile, int lineNum) { pThread->BeginForbidGC(szFile, lineNum); } }; #define TRIGGERSGC_NOSTOMP() do { \ Thread* curThread = GetThread(); \ if(curThread->GCNoTrigger()) \ { \ CONTRACT_ASSERT("TRIGGERSGC found in a GC_NOTRIGGER region.", Contract::GC_NoTrigger, Contract::GC_Mask, __FUNCTION__, __FILE__, __LINE__); \ } \ } while(0) #define TRIGGERSGC() do { \ TRIGGERSGC_NOSTOMP(); \ Thread::TriggersGC(GetThread()); \ } while(0) #else // ENABLE_CONTRACTS_IMPL #define BEGINFORBIDGC() #define ENDFORBIDGC() #define TRIGGERSGC_NOSTOMP() #define TRIGGERSGC() #endif // ENABLE_CONTRACTS_IMPL inline BOOL GC_ON_TRANSITIONS(BOOL val) { WRAPPER_NO_CONTRACT; #ifdef _DEBUG Thread* thread = GetThreadNULLOk(); if (thread == 0) return(FALSE); BOOL ret = thread->m_GCOnTransitionsOK; thread->m_GCOnTransitionsOK = val; return(ret); #else // _DEBUG return FALSE; #endif // !_DEBUG } #ifdef _DEBUG inline void ENABLESTRESSHEAP() { WRAPPER_NO_CONTRACT; Thread * thread = GetThreadNULLOk(); if (thread) { thread->EnableStressHeap(); } } void CleanStackForFastGCStress (); #define CLEANSTACKFORFASTGCSTRESS() \ if (g_pConfig->GetGCStressLevel() && g_pConfig->FastGCStressLevel() > 1) { \ CleanStackForFastGCStress (); \ } #else // _DEBUG #define CLEANSTACKFORFASTGCSTRESS() #endif // _DEBUG #ifdef _DEBUG_IMPL extern thread_local int t_ForbidGCLoaderUseCount; // Holder for incrementing the ForbidGCLoaderUse counter. class GCForbidLoaderUseHolder { public: GCForbidLoaderUseHolder() { WRAPPER_NO_CONTRACT; t_ForbidGCLoaderUseCount++; } ~GCForbidLoaderUseHolder() { WRAPPER_NO_CONTRACT; t_ForbidGCLoaderUseCount--; } }; #endif // Declaring this macro turns off the GC_TRIGGERS/THROWS contract in LoadTypeHandle. // If you do this, you must restrict your use of the loader only to retrieve TypeHandles // for types that have already been loaded and resolved. If you fail to observe this restriction, you will // reach a GC_TRIGGERS point somewhere in the loader and assert. If you're lucky, that is. // (If you're not lucky, you will introduce a GC hole.) // // The main user of this workaround is the GC stack crawl. It must parse signatures and retrieve // type handles for valuetypes in method parameters. Some other uses have creeped into the codebase - // some justified, others not. // // ENABLE_FORBID_GC_LOADER is *not* the same as using tokenNotToLoad to suppress loading. // You should use tokenNotToLoad in preference to ENABLE_FORBID. ENABLE_FORBID is a fragile // workaround and places enormous responsibilities on the caller. The only reason it exists at all // is that the GC stack crawl simply cannot tolerate exceptions or new GC's - that's an immovable // rock we're faced with. // // The key differences are: // // ENABLE_FORBID tokenNotToLoad // -------------------------------------------- ------------------------------------------------------ // caller must guarantee the type is already caller does not have to guarantee the type // loaded - otherwise, we will crash badly. is already loaded. // // loader will not throw, trigger gc or OOM loader may throw, trigger GC or OOM. // // // #ifdef ENABLE_CONTRACTS_IMPL #define ENABLE_FORBID_GC_LOADER_USE_IN_THIS_SCOPE() GCForbidLoaderUseHolder __gcfluh; \ CANNOTTHROWCOMPLUSEXCEPTION(); \ GCX_NOTRIGGER(); #else // _DEBUG_IMPL #define ENABLE_FORBID_GC_LOADER_USE_IN_THIS_SCOPE() ; #endif // _DEBUG_IMPL // This macro lets us define a conditional CONTRACT for the GC_TRIGGERS behavior. // This is for the benefit of a select group of callers that use the loader // in ForbidGC mode strictly to retrieve existing TypeHandles. The reason // we use a threadstate rather than an extra parameter is that these annoying // callers call the loader through intermediaries (MetaSig) and it proved to be too // cumbersome to pass this state down through all those callers. // // Don't make GC_TRIGGERS conditional just because your function ends up calling // LoadTypeHandle indirectly. We don't want to proliferate conditonal contracts more // than necessary so declare such functions as GC_TRIGGERS until the need // for the conditional contract is actually proven through code inspection or // coverage. #if defined(DACCESS_COMPILE) #define FORBIDGC_LOADER_USE_ENABLED() true #else // DACCESS_COMPILE #ifdef _DEBUG_IMPL #define FORBIDGC_LOADER_USE_ENABLED() (t_ForbidGCLoaderUseCount) #else // _DEBUG_IMPL // If you got an error about FORBIDGC_LOADER_USE_ENABLED being undefined, it's because you tried // to use this predicate in a free build outside of a CONTRACT or ASSERT. // #define FORBIDGC_LOADER_USE_ENABLED() (sizeof(YouCannotUseThisHere) != 0) #endif // _DEBUG_IMPL #endif // DACCESS_COMPILE // We have numerous places where we start up a managed thread. This includes several places in the // ThreadPool, the 'new Thread(...).Start()' case, and the Finalizer. Try to factor the code so our // base exception handling behavior is consistent across those places. The resulting code is convoluted, // but it's better than the prior situation of each thread being on a different plan. // If you add a new kind of managed thread (i.e. thread proc) to the system, you must: // // 1) Call HasStarted() before calling any ManagedThreadBase_* routine. // 2) Define a ManagedThreadBase_* routine for your scenario and declare it below. // 3) Always perform any AD transitions through the ManagedThreadBase_* mechanism. // 4) Allow the ManagedThreadBase_* mechanism to perform all your exception handling, including // dispatching of unhandled exception events, deciding what to swallow, etc. // 5) If you must separate your base thread proc behavior from your AD transitioning behavior, // define a second ManagedThreadADCall_* helper and declare it below. // 6) Never decide this is too much work and that you will roll your own thread proc code. // intentionally opaque. struct ManagedThreadCallState; struct ManagedThreadBase { static void KickOff(ADCallBackFcnType pTarget, LPVOID args); }; // DeadlockAwareLock is a base for building deadlock-aware locks. // Note that DeadlockAwareLock only works if ALL locks involved in the deadlock are deadlock aware. class DeadlockAwareLock { private: VolatilePtr m_pHoldingThread; #ifdef _DEBUG const char *m_description; #endif public: DeadlockAwareLock(const char *description = NULL); ~DeadlockAwareLock(); // Test for deadlock BOOL CanEnterLock(); // Call BeginEnterLock before attempting to acquire the lock BOOL TryBeginEnterLock(); // returns FALSE if deadlock void BeginEnterLock(); // Asserts if deadlock // Call EndEnterLock after acquiring the lock void EndEnterLock(); // Call LeaveLock after releasing the lock void LeaveLock(); const char *GetDescription(); private: CHECK CheckDeadlock(Thread *pThread); static void ReleaseBlockingLock() { Thread *pThread = GetThread(); pThread->m_pBlockingLock = NULL; } public: typedef StateHolder BlockingLockHolder; }; inline void SetTypeHandleOnThreadForAlloc(TypeHandle th) { // We are doing this unconditionally even though th is only used by ETW events in GC. When the ETW // event is not enabled we still need to set it because it may not be enabled here but by the // time we are checking in GC, the event is enabled - we don't want GC to read a random value // from before in this case. GetThread()->SetTHAllocContextObj(th); } class Compiler; // users of OFFSETOF__TLS__tls_CurrentThread macro expect the offset of these variables wrt to _tls_start to be stable. // Defining each of the following thread local variable separately without the struct causes the offsets to change in // different flavors of build. Eg. in chk build the offset of m_pThread is 0x4 while in ret build it becomes 0x8 as 0x4 is // occupied by m_pAddDomain. Packing all thread local variables in a struct and making struct instance to be thread local // ensures that the offsets of the variables are stable in all build flavors. struct ThreadLocalInfo { Thread* m_pThread; AppDomain* m_pAppDomain; // This field is read only by the SOS plugin to get the AppDomain void** m_EETlsData; // ClrTlsInfo::data }; class ThreadStateHolder { public: ThreadStateHolder (BOOL fNeed, DWORD state) { LIMITED_METHOD_CONTRACT; _ASSERTE (GetThreadNULLOk()); m_fNeed = fNeed; m_state = state; } ~ThreadStateHolder () { LIMITED_METHOD_CONTRACT; if (m_fNeed) { Thread *pThread = GetThread(); InterlockedAnd((LONG*)&pThread->m_State, ~m_state); } } private: BOOL m_fNeed; DWORD m_state; }; // Sets an NC threadstate if not already set, and restores the old state // of that bit upon destruction // fNeed > 0, make sure state is set, restored in destructor // fNeed = 0, no change // fNeed < 0, make sure state is reset, restored in destructor class ThreadStateNCStackHolder { public: ThreadStateNCStackHolder (BOOL fNeed, Thread::ThreadStateNoConcurrency state) { LIMITED_METHOD_CONTRACT; _ASSERTE (GetThreadNULLOk()); m_fNeed = fNeed; m_state = state; if (fNeed) { Thread *pThread = GetThread(); if (fNeed < 0) { // if the state is set, reset it if (pThread->HasThreadStateNC(state)) { pThread->ResetThreadStateNC(m_state); } else { m_fNeed = FALSE; } } else { // if the state is already set then no change is // necessary during the back out if(pThread->HasThreadStateNC(state)) { m_fNeed = FALSE; } else { pThread->SetThreadStateNC(state); } } } } ~ThreadStateNCStackHolder() { LIMITED_METHOD_CONTRACT; if (m_fNeed) { Thread *pThread = GetThread(); if (m_fNeed < 0) { pThread->SetThreadStateNC(m_state); // set it } else { pThread->ResetThreadStateNC(m_state); } } } private: BOOL m_fNeed; Thread::ThreadStateNoConcurrency m_state; }; BOOL Debug_IsLockedViaThreadSuspension(); inline BOOL IsWriteBarrierCopyEnabled() { #ifdef DACCESS_COMPILE return FALSE; #else // DACCESS_COMPILE return g_pConfig->IsWriteBarrierCopyEnabled(); #endif // DACCESS_COMPILE } BYTE* GetWriteBarrierCodeLocation(VOID* barrier); BOOL IsIPInWriteBarrierCodeCopy(PCODE controlPc); PCODE AdjustWriteBarrierIP(PCODE controlPc); #if !defined(DACCESS_COMPILE) extern thread_local Thread* t_pStackWalkerWalkingThread; #define SET_THREAD_TYPE_STACKWALKER(pThread) t_pStackWalkerWalkingThread = pThread #define CLEAR_THREAD_TYPE_STACKWALKER() t_pStackWalkerWalkingThread = NULL #else #define SET_THREAD_TYPE_STACKWALKER(pThread) #define CLEAR_THREAD_TYPE_STACKWALKER() #endif inline BOOL IsStackWalkerThread() { LIMITED_METHOD_CONTRACT; #if !defined(DACCESS_COMPILE) return t_pStackWalkerWalkingThread != NULL; #else return FALSE; #endif } class StackWalkerWalkingThreadHolder { public: StackWalkerWalkingThreadHolder(Thread* value) { LIMITED_METHOD_CONTRACT; #if !defined(DACCESS_COMPILE) m_PreviousValue = t_pStackWalkerWalkingThread; t_pStackWalkerWalkingThread = value; #endif } ~StackWalkerWalkingThreadHolder() { LIMITED_METHOD_CONTRACT; #if !defined(DACCESS_COMPILE) t_pStackWalkerWalkingThread = m_PreviousValue; #endif } private: Thread* m_PreviousValue; }; EXTERN_C Thread* GetThreadAsyncSafe(); #ifndef DACCESS_COMPILE #if defined(TARGET_WINDOWS) && defined(TARGET_AMD64) EXTERN_C void STDCALL ClrRestoreNonvolatileContextWorker(PCONTEXT ContextRecord, DWORD64 ssp); #endif void ClrRestoreNonvolatileContext(PCONTEXT ContextRecord, size_t targetSSP = 0); #endif // DACCESS_COMPILE #endif //__threads_h__