Make coreclr_initialize run cleanly for corewasmrun (#119217)

* Separate call invokers between JIT compatible interpreter
and interpreter with Portable EntryPoints.

* Handle native helpers with portable entry points

* Add simple interpreter managed frame stack walker when
native assert fires. This will help debugging asserts in interpreter
scenarios.

---------

Co-authored-by: Jan Kotas <[email protected]>

Author: AaronRobinsonMSFT

src/coreclr/debug/daccess/daccess.cpp

@@ -5635,7 +5635,7 @@ ClrDataAccess::GetJitHelperName(
 
         for (int i = 0; i < CORINFO_HELP_COUNT; i++)
         {
-            if (address == (TADDR)(pTable[i].pfnHelper))
+            if (address == pTable[i].pfnHelper)
                 return s_rgHelperNames[i];
         }
     }
@@ -5652,7 +5652,7 @@ ClrDataAccess::GetJitHelperName(
         PTR_READ(dac_cast<TADDR>(&hlpDynamicFuncTable), DYNAMIC_CORINFO_HELP_COUNT * sizeof(VMHELPDEF)));
     for (unsigned d = 0; d < DYNAMIC_CORINFO_HELP_COUNT; d++)
     {
-        if (address == (TADDR)(pDynamicTable[d].pfnHelper))
+        if (address == pDynamicTable[d].pfnHelper)
         {
             return s_rgHelperNames[s_rgDynamicHCallIds[d]];
         }
@@ -5991,7 +5991,7 @@ ClrDataAccess::GetMethodVarInfo(MethodDesc* methodDesc,
     BOOL success = DebugInfoManager::GetBoundariesAndVars(
         request,
         DebugInfoStoreNew, NULL, // allocator
-        BoundsType::Instrumented, 
+        BoundsType::Instrumented,
         NULL, NULL,
         &countNativeVarInfo, &nativeVars);
 

src/coreclr/debug/ee/controller.cpp

@@ -210,13 +210,13 @@ SharedPatchBypassBuffer* DebuggerControllerPatch::GetOrCreateSharedPatchBypassBu
         //
 
         BYTE* bufferBypassRW = pSharedPatchBypassBufferRW->BypassBuffer;
-        
+
         // Overwrite the *signed* displacement.
         int dwOldDisp = *(int*)(&patchBypassRX[instrAttrib.m_dwOffsetToDisp]);
         int dwNewDisp = offsetof(SharedPatchBypassBuffer, BypassBuffer) -
         (offsetof(SharedPatchBypassBuffer, PatchBypass) + instrAttrib.m_cbInstr);
         *(int*)(&patchBypassRW[instrAttrib.m_dwOffsetToDisp]) = dwNewDisp;
-        
+
         // This could be an LEA, which we'll just have to change into a MOV and copy the original address.
         if (((patchBypassRX[0] == 0x4C) || (patchBypassRX[0] == 0x48)) && (patchBypassRX[1] == 0x8d))
         {
@@ -229,7 +229,7 @@ SharedPatchBypassBuffer* DebuggerControllerPatch::GetOrCreateSharedPatchBypassBu
             _ASSERTE(instrAttrib.m_cOperandSize <= SharedPatchBypassBuffer::cbBufferBypass);
             // Copy the data into our buffer.
             memcpy(bufferBypassRW, this->address + instrAttrib.m_cbInstr + dwOldDisp, instrAttrib.m_cOperandSize);
-            
+
             if (instrAttrib.m_fIsWrite)
             {
                 // save the actual destination address and size so when we TriggerSingleStep() we can update the value
@@ -238,17 +238,17 @@ SharedPatchBypassBuffer* DebuggerControllerPatch::GetOrCreateSharedPatchBypassBu
             }
         }
     }
-    
+
     #endif // TARGET_AMD64
 
     m_pSharedPatchBypassBuffer->SetInstructionAttrib(instrAttrib);
-    
+
     // Since we just created a new buffer of code, but the CPU caches code and may
     // not be aware of our changes. This should force the CPU to dump any cached
     // instructions it has in this region and load the new ones from memory
     FlushInstructionCache(GetCurrentProcess(), patchBypassRW + CORDbg_BREAK_INSTRUCTION_SIZE,
     MAX_INSTRUCTION_LENGTH - CORDbg_BREAK_INSTRUCTION_SIZE);
-    
+
     return m_pSharedPatchBypassBuffer;
 }
 #endif // !FEATURE_EMULATE_SINGLESTEP
@@ -2405,29 +2405,29 @@ static bool _AddrIsJITHelper(PCODE addr)
     {
         for (int i = 0; i < CORINFO_HELP_COUNT; i++)
         {
-            if (hlpFuncTable[i].pfnHelper == (void*)addr)
+            if (hlpFuncTable[i].pfnHelper == addr)
             {
                 LOG((LF_CORDB, LL_INFO10000,
-                        "_ANIM: address of helper function found: 0x%08x\n",
+                        "_ANIM: address of helper function found: 0x%zx\n",
                         addr));
                 return true;
             }
         }
 
         for (unsigned d = 0; d < DYNAMIC_CORINFO_HELP_COUNT; d++)
         {
-            if (hlpDynamicFuncTable[d].pfnHelper == (void*)addr)
+            if (hlpDynamicFuncTable[d].pfnHelper == addr)
             {
                 LOG((LF_CORDB, LL_INFO10000,
-                        "_ANIM: address of helper function found: 0x%08x\n",
+                        "_ANIM: address of helper function found: 0x%zx\n",
                         addr));
                 return true;
             }
         }
 
         LOG((LF_CORDB, LL_INFO10000,
              "_ANIM: address within runtime dll, but not a helper function "
-             "0x%08x\n", addr));
+             "0x%zx\n", addr));
     }
 #else // !defined(HOST_64BIT) && !defined(TARGET_UNIX)
     // TODO: Figure out what we want to do here

src/coreclr/inc/regdisp.h

@@ -471,7 +471,9 @@ inline void FillContextPointers(PT_KNONVOLATILE_CONTEXT_POINTERS pCtxPtrs, PT_CO
     *(&pCtxPtrs->Tp) = &pCtx->Tp;
     *(&pCtxPtrs->Fp) = &pCtx->Fp;
     *(&pCtxPtrs->Ra) = &pCtx->Ra;
-#else // TARGET_RISCV64
+#elif defined(TARGET_WASM)
+    // Wasm doesn't have registers
+#else // TARGET_WASM
     PORTABILITY_ASSERT("FillContextPointers");
 #endif // _TARGET_???_ (ELSE)
 }

src/coreclr/pal/src/arch/wasm/stubs.cpp

@@ -8,9 +8,16 @@ SET_DEFAULT_DEBUG_CHANNEL(EXCEPT); // some headers have code with asserts, so do
 
 /* debugbreak */
 
+#ifdef _DEBUG
+extern void DBG_PrintInterpreterStack();
+#endif // _DEBUG
+
 extern "C" void
 DBG_DebugBreak()
 {
+#ifdef _DEBUG
+    DBG_PrintInterpreterStack();
+#endif // _DEBUG
     asm volatile ("unreachable");
 }
 

src/coreclr/vm/callhelpers.cpp

@@ -225,8 +225,8 @@ void* DispatchCallSimple(
     callDescrData.pTarget = pTargetAddress;
 
 #ifdef TARGET_WASM
-    PORTABILITY_ASSERT("wasm need to fill call description data");
-#endif
+    callDescrData.nArgsSize = numStackSlotsToCopy * sizeof(ARGHOLDER_TYPE);
+#endif // TARGET_WASM
 
     if ((dwDispatchCallSimpleFlags & DispatchCallSimple_CatchHandlerFoundNotification) != 0)
     {
@@ -519,9 +519,6 @@ void MethodDescCallSite::CallTargetWorker(const ARG_SLOT *pArguments, ARG_SLOT *
     CallDescrData callDescrData;
 
     callDescrData.pSrc = pTransitionBlock + sizeof(TransitionBlock);
-#ifdef TARGET_WASM
-    callDescrData.pTransitionBlock = (TransitionBlock*)pTransitionBlock;
-#endif
     _ASSERTE((nStackBytes % TARGET_POINTER_SIZE) == 0);
     callDescrData.numStackSlots = nStackBytes / TARGET_POINTER_SIZE;
 #ifdef CALLDESCR_ARGREGS
@@ -539,9 +536,8 @@ void MethodDescCallSite::CallTargetWorker(const ARG_SLOT *pArguments, ARG_SLOT *
     callDescrData.fpReturnSize = fpReturnSize;
     callDescrData.pTarget = m_pCallTarget;
 #ifdef TARGET_WASM
-    callDescrData.pMD = m_pMD;
     callDescrData.nArgsSize = m_argIt.GetArgSize();
-#endif
+#endif // TARGET_WASM
 
     CallDescrWorkerWithHandler(&callDescrData);
 

src/coreclr/vm/callhelpers.h

@@ -29,11 +29,8 @@ struct CallDescrData
     UINT32                      fpReturnSize;
     PCODE                       pTarget;
 #ifdef TARGET_WASM
-    // method description is used to compile the method with the interpreter
-    MethodDesc*                 pMD;
     // size of the arguments and the transition block are used to execute the method with the interpreter
     size_t                      nArgsSize;
-    TransitionBlock*            pTransitionBlock;
 #endif
 
 #ifdef CALLDESCR_RETBUFFARGREG

src/coreclr/vm/callingconvention.h

@@ -12,6 +12,10 @@
 #ifndef __CALLING_CONVENTION_INCLUDED
 #define __CALLING_CONVENTION_INCLUDED
 
+#ifdef FEATURE_INTERPRETER
+#include "../../interpreter/interpretershared.h"
+#endif // FEATURE_INTERPRETER
+
 BOOL IsRetBuffPassedAsFirstArg();
 
 // Describes how a single argument is laid out in registers and/or stack locations when given as an input to a
@@ -491,31 +495,31 @@ class ArgIteratorTemplate : public ARGITERATOR_BASE
 
                 while (typ == ELEMENT_TYPE_VALUETYPE &&
                     pMT->GetNumInstanceFields() == 1 && (!pMT->HasLayout()	||
-                    pMT->GetNumInstanceFieldBytes() == 4	
-                    )) // Don't do the optimization if we're getting specified anything but the trivial layout.	
-                {	
-                    FieldDesc * pFD = pMT->GetApproxFieldDescListRaw();	
+                    pMT->GetNumInstanceFieldBytes() == 4
+                    )) // Don't do the optimization if we're getting specified anything but the trivial layout.
+                {
+                    FieldDesc * pFD = pMT->GetApproxFieldDescListRaw();
                     CorElementType type = pFD->GetFieldType();
 
                     bool exitLoop = false;
-                    switch (type)	
+                    switch (type)
                     {
                         case ELEMENT_TYPE_VALUETYPE:
                         {
-                            //@todo: Is it more apropos to call LookupApproxFieldTypeHandle() here?	
-                            TypeHandle fldHnd = pFD->GetApproxFieldTypeHandleThrowing();	
+                            //@todo: Is it more apropos to call LookupApproxFieldTypeHandle() here?
+                            TypeHandle fldHnd = pFD->GetApproxFieldTypeHandleThrowing();
                             CONSISTENCY_CHECK(!fldHnd.IsNull());
                             pMT = fldHnd.GetMethodTable();
                             FALLTHROUGH;
-                        }	
+                        }
                         case ELEMENT_TYPE_PTR:
                         case ELEMENT_TYPE_I:
                         case ELEMENT_TYPE_U:
-                        case ELEMENT_TYPE_I4:	
+                        case ELEMENT_TYPE_I4:
                         case ELEMENT_TYPE_U4:
-                        {	
+                        {
                             typ = type;
-                            break;	
+                            break;
                         }
                         default:
                             exitLoop = true;
@@ -853,7 +857,7 @@ class ArgIteratorTemplate : public ARGITERATOR_BASE
             pLoc->m_idxFloatReg = floatRegOfsInBytes / FLOAT_REGISTER_SIZE;
             pLoc->m_cFloatReg = 1;
         }
-        else 
+        else
 #endif // UNIX_AMD64_ABI
         if (!TransitionBlock::IsStackArgumentOffset(argOffset))
         {
@@ -1874,11 +1878,11 @@ int ArgIteratorTemplate<ARGITERATOR_BASE>::GetNextOffset()
 
     return argOfs;
 #elif defined(TARGET_WASM)
-    int cbArg = ALIGN_UP(StackElemSize(argSize), TARGET_POINTER_SIZE);
+    int cbArg = ALIGN_UP(argSize, INTERP_STACK_SLOT_SIZE);
     int argOfs = TransitionBlock::GetOffsetOfArgs() + m_ofsStack;
-    
+
     m_ofsStack += cbArg;
-    
+
     return argOfs;
 #else
     PORTABILITY_ASSERT("ArgIteratorTemplate::GetNextOffset");

src/coreclr/vm/dllimport.cpp

@@ -5654,6 +5654,9 @@ PCODE PInvoke::GetStubForILStub(PInvokeMethodDesc* pNMD, MethodDesc** ppStubMD,
         // varargs goes through vararg PInvoke stub
         //
         pStub = TheVarargPInvokeStub(pNMD->HasRetBuffArg());
+
+        // Only vararg P/Invoke use shared stubs, they need a precode to push the hidden argument.
+        (void)pNMD->GetOrCreatePrecode();
     }
 
     if (pNMD->IsEarlyBound())