Move the thread alloc context off of Thread in CoreCLR (#103055)

Author: jkoritzinsky

src/coreclr/debug/daccess/dacdbiimpl.cpp

@@ -4713,8 +4713,16 @@ void DacDbiInterfaceImpl::GetThreadAllocInfo(VMPTR_Thread        vmThread,
 
     Thread * pThread = vmThread.GetDacPtr();
     gc_alloc_context* allocContext = pThread->GetAllocContext();
-    threadAllocInfo->m_allocBytesSOH = allocContext->alloc_bytes - (allocContext->alloc_limit - allocContext->alloc_ptr);
-    threadAllocInfo->m_allocBytesUOH = allocContext->alloc_bytes_uoh;
+    if (allocContext != nullptr)
+    {
+        threadAllocInfo->m_allocBytesSOH = allocContext->alloc_bytes - (allocContext->alloc_limit - allocContext->alloc_ptr);
+        threadAllocInfo->m_allocBytesUOH = allocContext->alloc_bytes_uoh;
+    }
+    else
+    {
+            threadAllocInfo->m_allocBytesSOH = 0;
+            threadAllocInfo->m_allocBytesUOH = 0;
+    }
 }
 
 // Set and reset the TSNC_DebuggerUserSuspend bit on the state of the specified thread

src/coreclr/debug/daccess/request.cpp

@@ -720,8 +720,18 @@ ClrDataAccess::GetThreadAllocData(CLRDATA_ADDRESS addr, struct DacpAllocData *da
 
     Thread* thread = PTR_Thread(TO_TADDR(addr));
 
-    data->allocBytes = TO_CDADDR(thread->m_alloc_context.alloc_bytes);
-    data->allocBytesLoh = TO_CDADDR(thread->m_alloc_context.alloc_bytes_uoh);
+    gc_alloc_context* pAllocContext = thread->GetAllocContext();
+
+    if (pAllocContext != NULL)
+    {
+        data->allocBytes = TO_CDADDR(pAllocContext->alloc_bytes);
+        data->allocBytesLoh = TO_CDADDR(pAllocContext->alloc_bytes_uoh);
+    }
+    else
+    {
+        data->allocBytes = TO_CDADDR(0);
+        data->allocBytesLoh = TO_CDADDR(0);
+    }
 
     SOSDacLeave();
     return hr;
@@ -816,8 +826,18 @@ HRESULT ClrDataAccess::GetThreadDataImpl(CLRDATA_ADDRESS threadAddr, struct Dacp
     threadData->osThreadId = (DWORD)thread->m_OSThreadId;
     threadData->state = thread->m_State;
     threadData->preemptiveGCDisabled = thread->m_fPreemptiveGCDisabled;
-    threadData->allocContextPtr = TO_CDADDR(thread->m_alloc_context.alloc_ptr);
-    threadData->allocContextLimit = TO_CDADDR(thread->m_alloc_context.alloc_limit);
+
+    gc_alloc_context* allocContext = thread->GetAllocContext();
+    if (allocContext)
+    {
+        threadData->allocContextPtr = TO_CDADDR(allocContext->alloc_ptr);
+        threadData->allocContextLimit = TO_CDADDR(allocContext->alloc_limit);
+    }
+    else
+    {
+        threadData->allocContextPtr = TO_CDADDR(0);
+        threadData->allocContextLimit = TO_CDADDR(0);
+    }
 
     threadData->fiberData = (CLRDATA_ADDRESS)NULL;
 

src/coreclr/vm/CMakeLists.txt

@@ -621,8 +621,8 @@ if(CLR_CMAKE_TARGET_ARCH_AMD64)
         ${ARCH_SOURCES_DIR}/GenericCLRToCOMCallStubs.asm
         ${ARCH_SOURCES_DIR}/getstate.asm
         ${ARCH_SOURCES_DIR}/JitHelpers_Fast.asm
+        ${ARCH_SOURCES_DIR}/JitHelpers_FastMP.asm
         ${ARCH_SOURCES_DIR}/JitHelpers_FastWriteBarriers.asm
-        ${ARCH_SOURCES_DIR}/JitHelpers_InlineGetThread.asm
         ${ARCH_SOURCES_DIR}/JitHelpers_SingleAppDomain.asm
         ${ARCH_SOURCES_DIR}/JitHelpers_Slow.asm
         ${ARCH_SOURCES_DIR}/patchedcode.asm

src/coreclr/vm/amd64/AsmMacros.inc

@@ -206,6 +206,26 @@ INLINE_GETTHREAD macro Reg
 
         endm
 
+;
+; Inlined macro to get the current thread's allocation context
+; Trashes rax and r11
+;
+
+INLINE_GET_ALLOC_CONTEXT macro Reg
+
+        EXTERN _tls_index: DWORD
+        EXTERN t_thread_alloc_context: DWORD
+
+        mov     r11d, [_tls_index]
+        mov     rax, gs:[OFFSET__TEB__ThreadLocalStoragePointer]
+        mov     rax, [rax + r11 * 8]
+        mov     r11d, SECTIONREL t_thread_alloc_context
+        add     rax, r11
+        mov     Reg, rax
+
+        endm
+
+
 ; if you change this code there will be corresponding code in JITInterfaceGen.cpp which will need to be changed
 ;
 

src/coreclr/vm/amd64/JitHelpers_FastMP.asm

@@ -0,0 +1,75 @@
+; Licensed to the .NET Foundation under one or more agreements.
+; The .NET Foundation licenses this file to you under the MIT license.
+
+; ***********************************************************************
+; File: JitHelpers_InlineGetThread.asm, see history in jithelp.asm
+;
+; ***********************************************************************
+
+include AsmMacros.inc
+include asmconstants.inc
+
+CopyValueClassUnchecked equ     ?CopyValueClassUnchecked@@YAXPEAX0PEAVMethodTable@@@Z
+JIT_Box                 equ     ?JIT_Box@@YAPEAVObject@@PEAUCORINFO_CLASS_STRUCT_@@PEAX@Z
+
+extern CopyValueClassUnchecked:proc
+extern JIT_Box:proc
+
+; HCIMPL2(Object*, JIT_Box, CORINFO_CLASS_HANDLE type, void* unboxedData)
+NESTED_ENTRY JIT_BoxFastMP, _TEXT
+
+        ; m_BaseSize is guaranteed to be a multiple of 8.
+        mov     r8d, [rcx + OFFSET__MethodTable__m_BaseSize]
+
+        INLINE_GET_ALLOC_CONTEXT r11
+        mov     r10, [r11 + OFFSETOF__gc_alloc_context__alloc_limit]
+        mov     rax, [r11 + OFFSETOF__gc_alloc_context__alloc_ptr]
+
+        add     r8, rax
+
+        cmp     r8, r10
+        ja      AllocFailed
+
+        test    rdx, rdx
+        je      NullRef
+
+        mov     [r11 + OFFSETOF__gc_alloc_context__alloc_ptr], r8
+        mov     [rax], rcx
+
+        ; Check whether the object contains pointers
+        test    dword ptr [rcx + OFFSETOF__MethodTable__m_dwFlags], MethodTable__enum_flag_ContainsPointers
+        jnz     ContainsPointers
+
+        ; We have no pointers - emit a simple inline copy loop
+        ; Copy the contents from the end
+        mov     ecx, [rcx + OFFSET__MethodTable__m_BaseSize]
+        sub     ecx, 18h  ; sizeof(ObjHeader) + sizeof(Object) + last slot
+
+align 16
+    CopyLoop:
+        mov     r8, [rdx+rcx]
+        mov     [rax+rcx+8], r8
+        sub     ecx, 8
+        jge     CopyLoop
+        REPRET
+
+    ContainsPointers:
+        ; Do call to CopyValueClassUnchecked(object, data, pMT)
+        push_vol_reg rax
+        alloc_stack 20h
+        END_PROLOGUE
+
+        mov     r8, rcx
+        lea     rcx, [rax + 8]
+        call    CopyValueClassUnchecked
+
+        add     rsp, 20h
+        pop     rax
+        ret
+
+    AllocFailed:
+    NullRef:
+        jmp     JIT_Box
+NESTED_END JIT_BoxFastMP, _TEXT
+
+        end