Fix for obscure scenario where InterpreterStub is invoked on an InterpMethod that does not already have a populated pCallStub

Author: kg

src/coreclr/vm/amd64/AsmHelpers.asm

@@ -5,6 +5,7 @@ include AsmMacros.inc
 include asmconstants.inc
 
 Thread_GetInterpThreadContext  TEXTEQU <?GetInterpThreadContext@Thread@@QEAAPEAUInterpThreadContext@@XZ>
+CreateNativeToInterpreterCallStub  TEXTEQU <?CreateNativeToInterpreterCallStub@@YAPEAUCallStubHeader@@PEAUInterpMethod@@@Z>
 
 extern PInvokeImportWorker:proc
 extern ThePreStub:proc
@@ -16,6 +17,7 @@ extern JIT_RareDisableHelperWorker:proc
 ifdef FEATURE_INTERPRETER
 extern ExecuteInterpretedMethod:proc
 extern Thread_GetInterpThreadContext:proc
+extern CreateNativeToInterpreterCallStub:proc
 endif
 
 extern g_pPollGC:QWORD
@@ -578,8 +580,34 @@ HaveInterpThreadContext:
         mov             r10, qword ptr [rax + OFFSETOF__InterpThreadContext__pStackPointer]
         ; Load the InterpMethod pointer from the IR bytecode
         mov             rax, qword ptr [rbx]
-        mov             rax, qword ptr [rax + OFFSETOF__InterpMethod__pCallStub]
-        lea             r11, qword ptr [rax + OFFSETOF__CallStubHeader__Routines]
+
+        ; Do we have a call stub? If not, create it.
+        mov             r11, qword ptr [rax + OFFSETOF__InterpMethod__pCallStub]
+        test            r11, r11
+        jnz             HaveCallStub
+
+        ; Pass the InterpMethod * to CreateNativeToInterpreterCallStub so it can populate the call stub
+        mov             rcx, rax
+        ; The x64 ABI considers the registers RAX, RCX, RDX, R8, R9, R10, R11, and XMM0-XMM5 volatile.
+        push            rax
+        push            rdx
+        push            r8
+        push            r9
+        push            r10
+        ; We're going to recreate r11 after this and we already trashed rcx so we don't need to save it.
+        call            CreateNativeToInterpreterCallStub
+        ; Restore the volatile registers
+        pop             r10
+        pop             r9
+        pop             r9
+        pop             rdx
+        pop             rax
+        ; CreateNativeToInterpreterCallStub populated pCallStub so now load it
+        mov             r11, qword ptr [rax + OFFSETOF__InterpMethod__pCallStub]
+
+HaveCallStub:
+
+        lea             r11, qword ptr [r11 + OFFSETOF__CallStubHeader__Routines]
         lea             rax, [rsp + __PWTB_TransitionBlock]
         ; Copy the arguments to the interpreter stack, invoke the InterpExecMethod and load the return value
         call            qword ptr [r11]