diff --git a/compiler-rt/lib/dfsan/dfsan.h b/compiler-rt/lib/dfsan/dfsan.h
--- a/compiler-rt/lib/dfsan/dfsan.h
+++ b/compiler-rt/lib/dfsan/dfsan.h
@@ -36,6 +36,12 @@
 // Zero out the TLS storage.
 void dfsan_clear_thread_local_state();
 
+// Set DFSan label and origin TLS of argument for a call.
+// Note that offset may not correspond with argument number.
+// Some arguments (aggregate/array) will use several offsets.
+void dfsan_set_arg_tls(uptr offset, dfsan_label label);
+void dfsan_set_arg_origin_tls(uptr offset, dfsan_origin o);
+
 // Return the origin associated with the first taint byte in the size bytes
 // from the address addr.
 dfsan_origin dfsan_read_origin_of_first_taint(const void *addr, uptr size);
diff --git a/compiler-rt/lib/dfsan/dfsan.cpp b/compiler-rt/lib/dfsan/dfsan.cpp
--- a/compiler-rt/lib/dfsan/dfsan.cpp
+++ b/compiler-rt/lib/dfsan/dfsan.cpp
@@ -937,6 +937,20 @@
   }
 }
 
+SANITIZER_INTERFACE_ATTRIBUTE
+void dfsan_set_arg_tls(uptr offset, dfsan_label label) {
+  // 2x to match ShadowTLSAlignment.
+  // ShadowTLSAlignment should probably be changed.
+  // TODO: Consider reducing ShadowTLSAlignment to 1.
+  // Aligning to 2 bytes is probably a remnant of fast16 mode.
+  ((dfsan_label *)__dfsan_arg_tls)[offset * 2] = label;
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void dfsan_set_arg_origin_tls(uptr offset, dfsan_origin o) {
+  __dfsan_arg_origin_tls[offset] = o;
+}
+
 extern "C" void dfsan_flush() {
   const uptr maxVirtualAddress = GetMaxUserVirtualAddress();
   for (unsigned i = 0; i < kMemoryLayoutSize; ++i) {
@@ -1106,7 +1120,7 @@
 
   dfsan_allocator_init();
 
-  DFsanThread *main_thread = DFsanThread::Create(nullptr, nullptr, nullptr);
+  DFsanThread *main_thread = DFsanThread::Create(nullptr, nullptr);
   SetCurrentThread(main_thread);
   main_thread->Init();
 
diff --git a/compiler-rt/lib/dfsan/dfsan_custom.cpp b/compiler-rt/lib/dfsan/dfsan_custom.cpp
--- a/compiler-rt/lib/dfsan/dfsan_custom.cpp
+++ b/compiler-rt/lib/dfsan/dfsan_custom.cpp
@@ -753,7 +753,6 @@
 }
 
 static int dfsan_pthread_create(pthread_t *thread, const pthread_attr_t *attr,
-                                void *start_routine_trampoline,
                                 void *start_routine, void *arg,
                                 dfsan_label *ret_label,
                                 bool track_origins = false) {
@@ -767,8 +766,7 @@
   AdjustStackSize((void *)(const_cast<pthread_attr_t *>(attr)));
 
   DFsanThread *t =
-      DFsanThread::Create(start_routine_trampoline,
-                          (thread_callback_t)start_routine, arg, track_origins);
+      DFsanThread::Create((thread_callback_t)start_routine, arg, track_origins);
   ScopedBlockSignals block(&t->starting_sigset_);
   int res = pthread_create(thread, attr, DFsanThreadStartFunc, t);
 
@@ -779,28 +777,22 @@
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE int __dfsw_pthread_create(
-    pthread_t *thread, const pthread_attr_t *attr,
-    void *(*start_routine_trampoline)(void *, void *, dfsan_label,
-                                      dfsan_label *),
-    void *start_routine, void *arg, dfsan_label thread_label,
-    dfsan_label attr_label, dfsan_label start_routine_label,
-    dfsan_label arg_label, dfsan_label *ret_label) {
-  return dfsan_pthread_create(thread, attr, (void *)start_routine_trampoline,
-                              start_routine, arg, ret_label);
+    pthread_t *thread, const pthread_attr_t *attr, void *start_routine,
+    void *arg, dfsan_label thread_label, dfsan_label attr_label,
+    dfsan_label start_routine_label, dfsan_label arg_label,
+    dfsan_label *ret_label) {
+  return dfsan_pthread_create(thread, attr, start_routine, arg, ret_label);
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE int __dfso_pthread_create(
-    pthread_t *thread, const pthread_attr_t *attr,
-    void *(*start_routine_trampoline)(void *, void *, dfsan_label,
-                                      dfsan_label *, dfsan_origin,
-                                      dfsan_origin *),
-    void *start_routine, void *arg, dfsan_label thread_label,
-    dfsan_label attr_label, dfsan_label start_routine_label,
-    dfsan_label arg_label, dfsan_label *ret_label, dfsan_origin thread_origin,
+    pthread_t *thread, const pthread_attr_t *attr, void *start_routine,
+    void *arg, dfsan_label thread_label, dfsan_label attr_label,
+    dfsan_label start_routine_label, dfsan_label arg_label,
+    dfsan_label *ret_label, dfsan_origin thread_origin,
     dfsan_origin attr_origin, dfsan_origin start_routine_origin,
     dfsan_origin arg_origin, dfsan_origin *ret_origin) {
-  return dfsan_pthread_create(thread, attr, (void *)start_routine_trampoline,
-                              start_routine, arg, ret_label, true);
+  return dfsan_pthread_create(thread, attr, start_routine, arg, ret_label,
+                              true);
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE int __dfsw_pthread_join(pthread_t thread,
@@ -825,22 +817,7 @@
 }
 
 struct dl_iterate_phdr_info {
-  int (*callback_trampoline)(void *callback, struct dl_phdr_info *info,
-                             size_t size, void *data, dfsan_label info_label,
-                             dfsan_label size_label, dfsan_label data_label,
-                             dfsan_label *ret_label);
-  void *callback;
-  void *data;
-};
-
-struct dl_iterate_phdr_origin_info {
-  int (*callback_trampoline)(void *callback, struct dl_phdr_info *info,
-                             size_t size, void *data, dfsan_label info_label,
-                             dfsan_label size_label, dfsan_label data_label,
-                             dfsan_label *ret_label, dfsan_origin info_origin,
-                             dfsan_origin size_origin, dfsan_origin data_origin,
-                             dfsan_origin *ret_origin);
-  void *callback;
+  int (*callback)(struct dl_phdr_info *info, size_t size, void *data);
   void *data;
 };
 
@@ -852,53 +829,28 @@
   dfsan_set_label(
       0, const_cast<char *>(reinterpret_cast<const char *>(info->dlpi_phdr)),
       sizeof(*info->dlpi_phdr) * info->dlpi_phnum);
-  dfsan_label ret_label;
-  return dipi->callback_trampoline(dipi->callback, info, size, dipi->data, 0, 0,
-                                   0, &ret_label);
-}
 
-int dl_iterate_phdr_origin_cb(struct dl_phdr_info *info, size_t size,
-                              void *data) {
-  dl_iterate_phdr_origin_info *dipi = (dl_iterate_phdr_origin_info *)data;
-  dfsan_set_label(0, *info);
-  dfsan_set_label(0, const_cast<char *>(info->dlpi_name),
-                  strlen(info->dlpi_name) + 1);
-  dfsan_set_label(
-      0, const_cast<char *>(reinterpret_cast<const char *>(info->dlpi_phdr)),
-      sizeof(*info->dlpi_phdr) * info->dlpi_phnum);
-  dfsan_label ret_label;
-  dfsan_origin ret_origin;
-  return dipi->callback_trampoline(dipi->callback, info, size, dipi->data, 0, 0,
-                                   0, &ret_label, 0, 0, 0, &ret_origin);
+  dfsan_clear_thread_local_state();
+  return dipi->callback(info, size, dipi->data);
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE int __dfsw_dl_iterate_phdr(
-    int (*callback_trampoline)(void *callback, struct dl_phdr_info *info,
-                               size_t size, void *data, dfsan_label info_label,
-                               dfsan_label size_label, dfsan_label data_label,
-                               dfsan_label *ret_label),
-    void *callback, void *data, dfsan_label callback_label,
-    dfsan_label data_label, dfsan_label *ret_label) {
-  dl_iterate_phdr_info dipi = { callback_trampoline, callback, data };
+    int (*callback)(struct dl_phdr_info *info, size_t size, void *data),
+    void *data, dfsan_label callback_label, dfsan_label data_label,
+    dfsan_label *ret_label) {
+  dl_iterate_phdr_info dipi = {callback, data};
   *ret_label = 0;
   return dl_iterate_phdr(dl_iterate_phdr_cb, &dipi);
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE int __dfso_dl_iterate_phdr(
-    int (*callback_trampoline)(void *callback, struct dl_phdr_info *info,
-                               size_t size, void *data, dfsan_label info_label,
-                               dfsan_label size_label, dfsan_label data_label,
-                               dfsan_label *ret_label, dfsan_origin info_origin,
-                               dfsan_origin size_origin,
-                               dfsan_origin data_origin,
-                               dfsan_origin *ret_origin),
-    void *callback, void *data, dfsan_label callback_label,
-    dfsan_label data_label, dfsan_label *ret_label,
-    dfsan_origin callback_origin, dfsan_origin data_origin,
-    dfsan_origin *ret_origin) {
-  dl_iterate_phdr_origin_info dipi = {callback_trampoline, callback, data};
+    int (*callback)(struct dl_phdr_info *info, size_t size, void *data),
+    void *data, dfsan_label callback_label, dfsan_label data_label,
+    dfsan_label *ret_label, dfsan_origin callback_origin,
+    dfsan_origin data_origin, dfsan_origin *ret_origin) {
+  dl_iterate_phdr_info dipi = {callback, data};
   *ret_label = 0;
-  return dl_iterate_phdr(dl_iterate_phdr_origin_cb, &dipi);
+  return dl_iterate_phdr(dl_iterate_phdr_cb, &dipi);
 }
 
 // This function is only available for glibc 2.27 or newer.  Mark it weak so
@@ -1629,10 +1581,7 @@
   SignalHandlerScope signal_handler_scope;
   ScopedClearThreadLocalState scoped_clear_tls;
 
-  // Clear shadows for all inputs provided by system. This is why DFSan
-  // instrumentation generates a trampoline function to each function pointer,
-  // and uses the trampoline to clear shadows. However sigaction does not use
-  // a function pointer directly, so we have to do this manually.
+  // Clear shadows for all inputs provided by system.
   dfsan_clear_arg_tls(0, sizeof(dfsan_label));
 
   typedef void (*signal_cb)(int x);
@@ -1733,22 +1682,18 @@
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE
-sighandler_t __dfsw_signal(int signum,
-                           void *(*handler_trampoline)(void *, int, dfsan_label,
-                                                       dfsan_label *),
-                           sighandler_t handler, dfsan_label signum_label,
-                           dfsan_label handler_label, dfsan_label *ret_label) {
+sighandler_t __dfsw_signal(int signum, sighandler_t handler,
+                           dfsan_label signum_label, dfsan_label handler_label,
+                           dfsan_label *ret_label) {
   return dfsan_signal(signum, handler, ret_label);
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE
-sighandler_t __dfso_signal(
-    int signum,
-    void *(*handler_trampoline)(void *, int, dfsan_label, dfsan_label *,
-                                dfsan_origin, dfsan_origin *),
-    sighandler_t handler, dfsan_label signum_label, dfsan_label handler_label,
-    dfsan_label *ret_label, dfsan_origin signum_origin,
-    dfsan_origin handler_origin, dfsan_origin *ret_origin) {
+sighandler_t __dfso_signal(int signum, sighandler_t handler,
+                           dfsan_label signum_label, dfsan_label handler_label,
+                           dfsan_label *ret_label, dfsan_origin signum_origin,
+                           dfsan_origin handler_origin,
+                           dfsan_origin *ret_origin) {
   return dfsan_signal(signum, handler, ret_label);
 }
 
@@ -2088,47 +2033,62 @@
                             addrlen_label, ret_label);
 }
 
-// Type of the trampoline function passed to the custom version of
-// dfsan_set_write_callback.
-typedef void (*write_trampoline_t)(
-    void *callback,
-    int fd, const void *buf, ssize_t count,
-    dfsan_label fd_label, dfsan_label buf_label, dfsan_label count_label);
-
-typedef void (*write_origin_trampoline_t)(
-    void *callback, int fd, const void *buf, ssize_t count,
-    dfsan_label fd_label, dfsan_label buf_label, dfsan_label count_label,
-    dfsan_origin fd_origin, dfsan_origin buf_origin, dfsan_origin count_origin);
+// Type of the function passed to dfsan_set_write_callback.
+typedef void (*write_dfsan_callback_t)(int fd, const void *buf, ssize_t count);
 
 // Calls to dfsan_set_write_callback() set the values in this struct.
 // Calls to the custom version of write() read (and invoke) them.
 static struct {
-  write_trampoline_t write_callback_trampoline = nullptr;
-  void *write_callback = nullptr;
+  write_dfsan_callback_t write_callback = nullptr;
 } write_callback_info;
 
-static struct {
-  write_origin_trampoline_t write_callback_trampoline = nullptr;
-  void *write_callback = nullptr;
-} write_origin_callback_info;
-
-SANITIZER_INTERFACE_ATTRIBUTE void
-__dfsw_dfsan_set_write_callback(
-    write_trampoline_t write_callback_trampoline,
-    void *write_callback,
-    dfsan_label write_callback_label,
+SANITIZER_INTERFACE_ATTRIBUTE void __dfsw_dfsan_set_write_callback(
+    write_dfsan_callback_t write_callback, dfsan_label write_callback_label,
     dfsan_label *ret_label) {
-  write_callback_info.write_callback_trampoline = write_callback_trampoline;
   write_callback_info.write_callback = write_callback;
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE void __dfso_dfsan_set_write_callback(
-    write_origin_trampoline_t write_callback_trampoline, void *write_callback,
-    dfsan_label write_callback_label, dfsan_label *ret_label,
-    dfsan_origin write_callback_origin, dfsan_origin *ret_origin) {
-  write_origin_callback_info.write_callback_trampoline =
-      write_callback_trampoline;
-  write_origin_callback_info.write_callback = write_callback;
+    write_dfsan_callback_t write_callback, dfsan_label write_callback_label,
+    dfsan_label *ret_label, dfsan_origin write_callback_origin,
+    dfsan_origin *ret_origin) {
+  write_callback_info.write_callback = write_callback;
+}
+
+static inline void setup_tls_args_for_write_callback(
+    dfsan_label fd_label, dfsan_label buf_label, dfsan_label count_label,
+    bool origins, dfsan_origin fd_origin, dfsan_origin buf_origin,
+    dfsan_origin count_origin) {
+  // The callback code will expect argument shadow labels in the args TLS,
+  // and origin labels in the origin args TLS.
+  // Previously this was done by a trampoline, but we want to remove this:
+  // https://github.com/llvm/llvm-project/issues/54172
+  //
+  // Instead, this code is manually setting up the args TLS data.
+  //
+  // The offsets used need to correspond with the instrumentation code,
+  // see llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp
+  // DFSanFunction::getShadowForTLSArgument.
+  // https://github.com/llvm/llvm-project/blob/0acc9e4b5edd8b39ff3d4c6d0e17f02007671c4e/llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp#L1684
+  // https://github.com/llvm/llvm-project/blob/0acc9e4b5edd8b39ff3d4c6d0e17f02007671c4e/llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp#L125
+  //
+  // Here the arguments are all primitives, but it can be more complex
+  // to compute offsets for array/aggregate type arguments.
+  //
+  // TODO(browneee): Consider a builtin to improve maintainabliity.
+  // With a builtin, we would provide the argument labels via builtin,
+  // and the builtin would reuse parts of the instrumentation code to ensure
+  // that this code and the instrumentation can never be out of sync.
+  // Note: Currently DFSan instrumentation does not run on this code, so
+  // the builtin may need to be handled outside DFSan instrumentation.
+  dfsan_set_arg_tls(0, fd_label);
+  dfsan_set_arg_tls(1, buf_label);
+  dfsan_set_arg_tls(2, count_label);
+  if (origins) {
+    dfsan_set_arg_origin_tls(0, fd_origin);
+    dfsan_set_arg_origin_tls(1, buf_origin);
+    dfsan_set_arg_origin_tls(2, count_origin);
+  }
 }
 
 SANITIZER_INTERFACE_ATTRIBUTE int
@@ -2136,10 +2096,9 @@
              dfsan_label fd_label, dfsan_label buf_label,
              dfsan_label count_label, dfsan_label *ret_label) {
   if (write_callback_info.write_callback) {
-    write_callback_info.write_callback_trampoline(
-        write_callback_info.write_callback,
-        fd, buf, count,
-        fd_label, buf_label, count_label);
+    setup_tls_args_for_write_callback(fd_label, buf_label, count_label, false,
+                                      0, 0, 0);
+    write_callback_info.write_callback(fd, buf, count);
   }
 
   *ret_label = 0;
@@ -2151,10 +2110,10 @@
     dfsan_label buf_label, dfsan_label count_label, dfsan_label *ret_label,
     dfsan_origin fd_origin, dfsan_origin buf_origin, dfsan_origin count_origin,
     dfsan_origin *ret_origin) {
-  if (write_origin_callback_info.write_callback) {
-    write_origin_callback_info.write_callback_trampoline(
-        write_origin_callback_info.write_callback, fd, buf, count, fd_label,
-        buf_label, count_label, fd_origin, buf_origin, count_origin);
+  if (write_callback_info.write_callback) {
+    setup_tls_args_for_write_callback(fd_label, buf_label, count_label, true,
+                                      fd_origin, buf_origin, count_origin);
+    write_callback_info.write_callback(fd, buf, count);
   }
 
   *ret_label = 0;
diff --git a/compiler-rt/lib/dfsan/dfsan_thread.h b/compiler-rt/lib/dfsan/dfsan_thread.h
--- a/compiler-rt/lib/dfsan/dfsan_thread.h
+++ b/compiler-rt/lib/dfsan/dfsan_thread.h
@@ -24,8 +24,7 @@
   // NOTE: There is no DFsanThread constructor. It is allocated
   // via mmap() and *must* be valid in zero-initialized state.
 
-  static DFsanThread *Create(void *start_routine_trampoline,
-                             thread_callback_t start_routine, void *arg,
+  static DFsanThread *Create(thread_callback_t start_routine, void *arg,
                              bool track_origins = false);
   static void TSDDtor(void *tsd);
   void Destroy();
@@ -59,7 +58,6 @@
 
   bool AddrIsInStack(uptr addr);
 
-  void *start_routine_trampoline_;
   thread_callback_t start_routine_;
   void *arg_;
   bool track_origins_;
diff --git a/compiler-rt/lib/dfsan/dfsan_thread.cpp b/compiler-rt/lib/dfsan/dfsan_thread.cpp
--- a/compiler-rt/lib/dfsan/dfsan_thread.cpp
+++ b/compiler-rt/lib/dfsan/dfsan_thread.cpp
@@ -7,13 +7,11 @@
 
 namespace __dfsan {
 
-DFsanThread *DFsanThread::Create(void *start_routine_trampoline,
-                                 thread_callback_t start_routine, void *arg,
+DFsanThread *DFsanThread::Create(thread_callback_t start_routine, void *arg,
                                  bool track_origins) {
   uptr PageSize = GetPageSizeCached();
   uptr size = RoundUpTo(sizeof(DFsanThread), PageSize);
   DFsanThread *thread = (DFsanThread *)MmapOrDie(size, __func__);
-  thread->start_routine_trampoline_ = start_routine_trampoline;
   thread->start_routine_ = start_routine;
   thread->arg_ = arg;
   thread->track_origins_ = track_origins;
@@ -74,23 +72,15 @@
     return 0;
   }
 
-  CHECK(start_routine_trampoline_);
+  // The only argument is void* arg.
+  //
+  // We have never supported propagating the pointer arg as tainted,
+  // __dfsw_pthread_create/__dfso_pthread_create ignore the taint label.
+  // Note that the bytes pointed-to (probably the much more common case)
+  // can still have taint labels attached to them.
+  dfsan_clear_thread_local_state();
 
-  typedef void *(*thread_callback_trampoline_t)(void *, void *, dfsan_label,
-                                                dfsan_label *);
-  typedef void *(*thread_callback_origin_trampoline_t)(
-      void *, void *, dfsan_label, dfsan_label *, dfsan_origin, dfsan_origin *);
-
-  dfsan_label ret_label;
-  if (!track_origins_)
-    return ((thread_callback_trampoline_t)
-                start_routine_trampoline_)((void *)start_routine_, arg_, 0,
-                                           &ret_label);
-
-  dfsan_origin ret_origin;
-  return ((thread_callback_origin_trampoline_t)
-              start_routine_trampoline_)((void *)start_routine_, arg_, 0,
-                                         &ret_label, 0, &ret_origin);
+  return start_routine_(arg_);
 }
 
 DFsanThread::StackBounds DFsanThread::GetStackBounds() const {
diff --git a/llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp
--- a/llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp
@@ -480,14 +480,12 @@
   bool isInstrumented(const Function *F);
   bool isInstrumented(const GlobalAlias *GA);
   bool isForceZeroLabels(const Function *F);
-  FunctionType *getTrampolineFunctionType(FunctionType *T);
   TransformedFunction getCustomFunctionType(FunctionType *T);
   WrapperKind getWrapperKind(Function *F);
   void addGlobalNameSuffix(GlobalValue *GV);
   Function *buildWrapperFunction(Function *F, StringRef NewFName,
                                  GlobalValue::LinkageTypes NewFLink,
                                  FunctionType *NewFT);
-  Constant *getOrBuildTrampolineFunction(FunctionType *FT, StringRef FName);
   void initializeCallbackFunctions(Module &M);
   void initializeRuntimeFunctions(Module &M);
   void injectMetadataGlobals(Module &M);
@@ -794,25 +792,6 @@
       SpecialCaseList::createOrDie(AllABIListFiles, *vfs::getRealFileSystem()));
 }
 
-FunctionType *DataFlowSanitizer::getTrampolineFunctionType(FunctionType *T) {
-  assert(!T->isVarArg());
-  SmallVector<Type *, 4> ArgTypes;
-  ArgTypes.push_back(T->getPointerTo());
-  ArgTypes.append(T->param_begin(), T->param_end());
-  ArgTypes.append(T->getNumParams(), PrimitiveShadowTy);
-  Type *RetType = T->getReturnType();
-  if (!RetType->isVoidTy())
-    ArgTypes.push_back(PrimitiveShadowPtrTy);
-
-  if (shouldTrackOrigins()) {
-    ArgTypes.append(T->getNumParams(), OriginTy);
-    if (!RetType->isVoidTy())
-      ArgTypes.push_back(OriginPtrTy);
-  }
-
-  return FunctionType::get(T->getReturnType(), ArgTypes, false);
-}
-
 TransformedFunction DataFlowSanitizer::getCustomFunctionType(FunctionType *T) {
   SmallVector<Type *, 4> ArgTypes;
 
@@ -824,15 +803,8 @@
   for (unsigned I = 0, E = T->getNumParams(); I != E; ++I) {
     Type *ParamType = T->getParamType(I);
     FunctionType *FT;
-    if (isa<PointerType>(ParamType) &&
-        (FT = dyn_cast<FunctionType>(ParamType->getPointerElementType()))) {
-      ArgumentIndexMapping.push_back(ArgTypes.size());
-      ArgTypes.push_back(getTrampolineFunctionType(FT)->getPointerTo());
-      ArgTypes.push_back(Type::getInt8PtrTy(*Ctx));
-    } else {
-      ArgumentIndexMapping.push_back(ArgTypes.size());
-      ArgTypes.push_back(ParamType);
-    }
+    ArgumentIndexMapping.push_back(ArgTypes.size());
+    ArgTypes.push_back(ParamType);
   }
   for (unsigned I = 0, E = T->getNumParams(); I != E; ++I)
     ArgTypes.push_back(PrimitiveShadowTy);
@@ -1181,61 +1153,6 @@
   return NewF;
 }
 
-Constant *DataFlowSanitizer::getOrBuildTrampolineFunction(FunctionType *FT,
-                                                          StringRef FName) {
-  FunctionType *FTT = getTrampolineFunctionType(FT);
-  FunctionCallee C = Mod->getOrInsertFunction(FName, FTT);
-  Function *F = dyn_cast<Function>(C.getCallee());
-  if (F && F->isDeclaration()) {
-    F->setLinkage(GlobalValue::LinkOnceODRLinkage);
-    BasicBlock *BB = BasicBlock::Create(*Ctx, "entry", F);
-    std::vector<Value *> Args;
-    Function::arg_iterator AI = F->arg_begin() + 1;
-    for (unsigned N = FT->getNumParams(); N != 0; ++AI, --N)
-      Args.push_back(&*AI);
-    CallInst *CI = CallInst::Create(FT, &*F->arg_begin(), Args, "", BB);
-    Type *RetType = FT->getReturnType();
-    ReturnInst *RI = RetType->isVoidTy() ? ReturnInst::Create(*Ctx, BB)
-                                         : ReturnInst::Create(*Ctx, CI, BB);
-
-    // F is called by a wrapped custom function with primitive shadows. So
-    // its arguments and return value need conversion.
-    DFSanFunction DFSF(*this, F, /*IsNativeABI=*/true,
-                       /*IsForceZeroLabels=*/false);
-    Function::arg_iterator ValAI = F->arg_begin(), ShadowAI = AI;
-    ++ValAI;
-    for (unsigned N = FT->getNumParams(); N != 0; ++ValAI, ++ShadowAI, --N) {
-      Value *Shadow =
-          DFSF.expandFromPrimitiveShadow(ValAI->getType(), &*ShadowAI, CI);
-      DFSF.ValShadowMap[&*ValAI] = Shadow;
-    }
-    Function::arg_iterator RetShadowAI = ShadowAI;
-    const bool ShouldTrackOrigins = shouldTrackOrigins();
-    if (ShouldTrackOrigins) {
-      ValAI = F->arg_begin();
-      ++ValAI;
-      Function::arg_iterator OriginAI = ShadowAI;
-      if (!RetType->isVoidTy())
-        ++OriginAI;
-      for (unsigned N = FT->getNumParams(); N != 0; ++ValAI, ++OriginAI, --N) {
-        DFSF.ValOriginMap[&*ValAI] = &*OriginAI;
-      }
-    }
-    DFSanVisitor(DFSF).visitCallInst(*CI);
-    if (!RetType->isVoidTy()) {
-      Value *PrimitiveShadow = DFSF.collapseToPrimitiveShadow(
-          DFSF.getShadow(RI->getReturnValue()), RI);
-      new StoreInst(PrimitiveShadow, &*RetShadowAI, RI);
-      if (ShouldTrackOrigins) {
-        Value *Origin = DFSF.getOrigin(RI->getReturnValue());
-        new StoreInst(Origin, &*std::prev(F->arg_end()), RI);
-      }
-    }
-  }
-
-  return cast<Constant>(C.getCallee());
-}
-
 // Initialize DataFlowSanitizer runtime functions and declare them in the module
 void DataFlowSanitizer::initializeRuntimeFunctions(Module &M) {
   {
@@ -2907,20 +2824,7 @@
     for (unsigned N = FT->getNumParams(); N != 0; ++I, --N) {
       Type *T = (*I)->getType();
       FunctionType *ParamFT;
-      if (isa<PointerType>(T) &&
-          (ParamFT = dyn_cast<FunctionType>(T->getPointerElementType()))) {
-        std::string TName = "dfst";
-        TName += utostr(FT->getNumParams() - N);
-        TName += "$";
-        TName += F.getName();
-        Constant *Trampoline =
-            DFSF.DFS.getOrBuildTrampolineFunction(ParamFT, TName);
-        Args.push_back(Trampoline);
-        Args.push_back(
-            IRB.CreateBitCast(*I, Type::getInt8PtrTy(*DFSF.DFS.Ctx)));
-      } else {
-        Args.push_back(*I);
-      }
+      Args.push_back(*I);
     }
 
     // Adds shadow arguments.
diff --git a/llvm/test/Instrumentation/DataFlowSanitizer/abilist.ll b/llvm/test/Instrumentation/DataFlowSanitizer/abilist.ll
--- a/llvm/test/Instrumentation/DataFlowSanitizer/abilist.ll
+++ b/llvm/test/Instrumentation/DataFlowSanitizer/abilist.ll
@@ -42,7 +42,7 @@
   ; CHECK: call i32 @__dfsw_custom2(i32 1, i32 2, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext 0, i[[#SBITS]]* %[[LABELRETURN]])
   call i32 @custom2(i32 1, i32 2)
 
-  ; CHECK: call void @__dfsw_customcb({{.*}} @"dfst0$customcb", i8* bitcast ({{.*}} @cb.dfsan to i8*), i[[#SBITS]] zeroext 0)
+  ; CHECK: call void @__dfsw_customcb({{.*}} @cb.dfsan, i[[#SBITS]] zeroext 0)
   call void @customcb(i32 (i32)* @cb)
 
   ; CHECK: %[[LABELVA1_0:.*]] = getelementptr inbounds [2 x i[[#SBITS]]], [2 x i[[#SBITS]]]* %[[LABELVA1]], i32 0, i32 0
@@ -93,12 +93,5 @@
 ; CHECK: declare void @__dfsw_custom1(i32, i32, i[[#SBITS]], i[[#SBITS]])
 ; CHECK: declare i32 @__dfsw_custom2(i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*)
 
-; CHECK-LABEL: define linkonce_odr i32 @"dfst0$customcb"
-; CHECK-SAME: (i32 (i32)* %0, i32 %1, i[[#SBITS]] %2, i[[#SBITS]]* %3)
-; CHECK: %[[CALL:.*]] = call i32 %0(i32 %1)
-; CHECK: %[[RVSHADOW2:.*]] = load i[[#SBITS]], {{.*}} @__dfsan_retval_tls
-; CHECK: store i[[#SBITS]] %[[RVSHADOW2]], i[[#SBITS]]* %3
-; CHECK: ret i32 %[[CALL]]
-
 ; CHECK: declare void @__dfsw_custom3(i32, i[[#SBITS]], i[[#SBITS]]*, ...)
 ; CHECK: declare i32 @__dfsw_custom4(i32, i[[#SBITS]], i[[#SBITS]]*, i[[#SBITS]]*, ...)
diff --git a/llvm/test/Instrumentation/DataFlowSanitizer/abilist_aggregate.ll b/llvm/test/Instrumentation/DataFlowSanitizer/abilist_aggregate.ll
--- a/llvm/test/Instrumentation/DataFlowSanitizer/abilist_aggregate.ll
+++ b/llvm/test/Instrumentation/DataFlowSanitizer/abilist_aggregate.ll
@@ -137,7 +137,7 @@
   ; CHECK: [[B0:%.*]] = extractvalue [2 x i[[#SBITS]]] [[B]], 0
   ; CHECK: [[B1:%.*]] = extractvalue [2 x i[[#SBITS]]] [[B]], 1
   ; CHECK: [[B01:%.*]] = or i[[#SBITS]] [[B0]], [[B1]]
-  ; CHECK: [[R:%.*]]  = call { i1, i7 } @__dfsw_custom_cb({ i1, i7 } ({ i1, i7 } ({ i32, i1 }, [2 x i7])*, { i32, i1 }, [2 x i7], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*)* @"dfst0$custom_cb", i8* bitcast ({ i1, i7 } ({ i32, i1 }, [2 x i7])* @cb.dfsan to i8*), { i32, i1 } %a, [2 x i7] %b, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext [[A01]], i[[#SBITS]] zeroext [[B01]], i[[#SBITS]]* %labelreturn)
+  ; CHECK: [[R:%.*]]  = call { i1, i7 } @__dfsw_custom_cb({ i1, i7 } ({ i32, i1 }, [2 x i7])* @cb.dfsan, { i32, i1 } %a, [2 x i7] %b, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext [[A01]], i[[#SBITS]] zeroext [[B01]], i[[#SBITS]]* %labelreturn)
   ; CHECK: [[RE:%.*]] = load i[[#SBITS]], i[[#SBITS]]* %labelreturn, align [[#SBYTES]]
   ; CHECK: [[RS0:%.*]] = insertvalue { i[[#SBITS]], i[[#SBITS]] } undef, i[[#SBITS]] [[RE]], 0
   ; CHECK: [[RS1:%.*]] = insertvalue { i[[#SBITS]], i[[#SBITS]] } [[RS0]], i[[#SBITS]] [[RE]], 1
@@ -186,20 +186,18 @@
 ; CHECK: [[B:%.*]] = load [2 x i[[#SBITS]]], [2 x i[[#SBITS]]]* inttoptr (i64 add (i64 ptrtoint ([100 x i64]* @__dfsan_arg_tls to i64), i64 [[#mul(2,SBYTES) + max(SBYTES,2)]]) to [2 x i[[#SBITS]]]*), align [[ALIGN:2]]
 ; CHECK: [[A:%.*]] = load { i[[#SBITS]], i[[#SBITS]] }, { i[[#SBITS]], i[[#SBITS]] }* inttoptr (i64 add (i64 ptrtoint ([100 x i64]* @__dfsan_arg_tls to i64), i64 2) to { i[[#SBITS]], i[[#SBITS]] }*), align [[ALIGN]]
 ; CHECK: [[CB:%.*]] = load i[[#SBITS]], i[[#SBITS]]* bitcast ([100 x i64]* @__dfsan_arg_tls to i[[#SBITS]]*), align [[ALIGN]]
-; CHECK: [[CAST:%.*]] = bitcast { i1, i7 } ({ i32, i1 }, [2 x i7])* %0 to i8*
 ; CHECK: [[A0:%.*]] = extractvalue { i[[#SBITS]], i[[#SBITS]] } [[A]], 0
 ; CHECK: [[A1:%.*]] = extractvalue { i[[#SBITS]], i[[#SBITS]] } [[A]], 1
 ; CHECK: [[A01:%.*]] = or i[[#SBITS]] [[A0]], [[A1]]
 ; CHECK: [[B0:%.*]] = extractvalue [2 x i[[#SBITS]]] [[B]], 0
 ; CHECK: [[B1:%.*]] = extractvalue [2 x i[[#SBITS]]] [[B]], 1
 ; CHECK: [[B01:%.*]] = or i[[#SBITS]] [[B0]], [[B1]]
-; CHECK: [[R:%.*]]  = call { i1, i7 } @__dfsw_custom_cb({ i1, i7 } ({ i1, i7 } ({ i32, i1 }, [2 x i7])*, { i32, i1 }, [2 x i7], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*)* @"dfst0$custom_cb", i8* [[CAST]], { i32, i1 } %1, [2 x i7] %2, i[[#SBITS]] zeroext [[CB]], i[[#SBITS]] zeroext [[A01]], i[[#SBITS]] zeroext [[B01]], i[[#SBITS]]* %labelreturn)
+; CHECK: [[R:%.*]]  = call { i1, i7 } @__dfsw_custom_cb({ i1, i7 } ({ i32, i1 }, [2 x i7])* %0, { i32, i1 } %1, [2 x i7] %2, i[[#SBITS]] zeroext [[CB]], i[[#SBITS]] zeroext [[A01]], i[[#SBITS]] zeroext [[B01]], i[[#SBITS]]* %labelreturn)
 ; CHECK: [[RE:%.*]] = load i[[#SBITS]], i[[#SBITS]]* %labelreturn, align [[#SBYTES]]
 ; CHECK: [[RS0:%.*]] = insertvalue { i[[#SBITS]], i[[#SBITS]] } undef, i[[#SBITS]] [[RE]], 0
 ; CHECK: [[RS1:%.*]] = insertvalue { i[[#SBITS]], i[[#SBITS]] } [[RS0]], i[[#SBITS]] [[RE]], 1
 ; CHECK: store { i[[#SBITS]], i[[#SBITS]] } [[RS1]], { i[[#SBITS]], i[[#SBITS]] }* bitcast ([100 x i64]* @__dfsan_retval_tls to { i[[#SBITS]], i[[#SBITS]] }*), align [[ALIGN]]
 
-
 define {i1, i7} @custom_with_ret({i32, i1} %a, [2 x i7] %b) {
   ; CHECK: define linkonce_odr { i1, i7 } @"dfsw$custom_with_ret"({ i32, i1 } %0, [2 x i7] %1)
   ; CHECK: %labelreturn = alloca i[[#SBITS]], align [[#SBYTES]]
@@ -250,19 +248,4 @@
 ; CHECK: declare void @__dfsw_custom_without_ret({ i32, i1 }, [2 x i7], i[[#SBITS]], i[[#SBITS]])
 ; CHECK: declare void @__dfsw_custom_varg({ i32, i1 }, i[[#SBITS]], i[[#SBITS]]*, ...)
 
-; CHECK: declare { i1, i7 } @__dfsw_custom_cb({ i1, i7 } ({ i1, i7 } ({ i32, i1 }, [2 x i7])*, { i32, i1 }, [2 x i7], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*)*, i8*, { i32, i1 }, [2 x i7], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*)
-
-; CHECK: define linkonce_odr { i1, i7 } @"dfst0$custom_cb"({ i1, i7 } ({ i32, i1 }, [2 x i7])* %0, { i32, i1 } %1, [2 x i7] %2, i[[#SBITS]] %3, i[[#SBITS]] %4, i[[#SBITS]]* %5) {
-; CHECK: [[A0:%.*]] = insertvalue { i[[#SBITS]], i[[#SBITS]] } undef, i[[#SBITS]] %3, 0
-; CHECK: [[A1:%.*]] = insertvalue { i[[#SBITS]], i[[#SBITS]] } [[A0]], i[[#SBITS]] %3, 1
-; CHECK: [[B0:%.*]] = insertvalue [2 x i[[#SBITS]]] undef, i[[#SBITS]] %4, 0
-; CHECK: [[B1:%.*]] = insertvalue [2 x i[[#SBITS]]] [[B0]], i[[#SBITS]] %4, 1
-; CHECK: store { i[[#SBITS]], i[[#SBITS]] } [[A1]], { i[[#SBITS]], i[[#SBITS]] }* bitcast ([100 x i64]* @__dfsan_arg_tls to { i[[#SBITS]], i[[#SBITS]] }*), align [[ALIGN:2]]
-; CHECK: store [2 x i[[#SBITS]]] [[B1]], [2 x i[[#SBITS]]]* inttoptr (i64 add (i64 ptrtoint ([100 x i64]* @__dfsan_arg_tls to i64), i64 [[#mul(2,SBYTES)]]) to [2 x i[[#SBITS]]]*), align [[ALIGN]]
-; CHECK: [[R:%.*]] = call { i1, i7 } %0({ i32, i1 } %1, [2 x i7] %2)
-; CHECK: %_dfsret = load { i[[#SBITS]], i[[#SBITS]] }, { i[[#SBITS]], i[[#SBITS]] }* bitcast ([100 x i64]* @__dfsan_retval_tls to { i[[#SBITS]], i[[#SBITS]] }*), align [[ALIGN]]
-; CHECK: [[RE0:%.*]] = extractvalue { i[[#SBITS]], i[[#SBITS]] } %_dfsret, 0
-; CHECK: [[RE1:%.*]] = extractvalue { i[[#SBITS]], i[[#SBITS]] } %_dfsret, 1
-; CHECK: [[RE01:%.*]] = or i[[#SBITS]] [[RE0]], [[RE1]]
-; CHECK: store i[[#SBITS]] [[RE01]], i[[#SBITS]]* %5, align [[#SBYTES]]
-; CHECK: ret { i1, i7 } [[R]]
+; CHECK: declare { i1, i7 } @__dfsw_custom_cb({ i1, i7 } ({ i32, i1 }, [2 x i7])*, { i32, i1 }, [2 x i7], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*)
diff --git a/llvm/test/Instrumentation/DataFlowSanitizer/custom_fun_callback_attributes.ll b/llvm/test/Instrumentation/DataFlowSanitizer/custom_fun_callback_attributes.ll
--- a/llvm/test/Instrumentation/DataFlowSanitizer/custom_fun_callback_attributes.ll
+++ b/llvm/test/Instrumentation/DataFlowSanitizer/custom_fun_callback_attributes.ll
@@ -16,16 +16,13 @@
 define void @call_custom_funs_with_callbacks(i8 (i32, double)* %callback_arg) {
   ;; The callback should have attribute 'nonnull':
   ; CHECK: call signext i32 @__dfsw_custom_fun_one_callback(
-  ; CHECK: nonnull @"dfst0$custom_fun_one_callback"
   %call1 = call signext i32 @custom_fun_one_callback(
     i8 (i32, double)* nonnull @a_callback_fun
   )
 
   ;; Call a custom function with two callbacks.  Check their annotations.
   ; CHECK: call i32 @__dfsw_custom_fun_two_callbacks(
-  ; CHECK: nonnull @"dfst0$custom_fun_two_callbacks"
   ; CHECK: i64 12345
-  ; CHECK: noalias @"dfst2$custom_fun_two_callbacks"
   %call2 = call i32 @custom_fun_two_callbacks(
     i8 (i32, double)* nonnull @a_callback_fun,
     i64 12345,
diff --git a/llvm/test/Instrumentation/DataFlowSanitizer/origin_abilist.ll b/llvm/test/Instrumentation/DataFlowSanitizer/origin_abilist.ll
--- a/llvm/test/Instrumentation/DataFlowSanitizer/origin_abilist.ll
+++ b/llvm/test/Instrumentation/DataFlowSanitizer/origin_abilist.ll
@@ -199,7 +199,7 @@
   ; CHECK: %labelreturn = alloca i[[#SBITS]], align [[#SBYTES]]
   ; CHECK: [[BS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 2) to i[[#SBITS]]*), align 2
   ; CHECK: [[AS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_arg_tls to i[[#SBITS]]*), align 2
-  ; CHECK: {{.*}} = call i32 @__dfso_custom_cb_with_ret(i32 (i32 (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32*)* @"dfst0$custom_cb_with_ret", i8* bitcast (i32 (i32, i32)* @cb_with_ret.dfsan to i8*), i32 %a, i32 %b, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i[[#SBITS]]* %labelreturn, i32 zeroext 0, i32 zeroext [[AO]], i32 zeroext [[BO]], i32* %originreturn)
+  ; CHECK: {{.*}} = call i32 @__dfso_custom_cb_with_ret(i32 (i32, i32)* @cb_with_ret.dfsan, i32 %a, i32 %b, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i[[#SBITS]]* %labelreturn, i32 zeroext 0, i32 zeroext [[AO]], i32 zeroext [[BO]], i32* %originreturn)
   ; CHECK: [[RS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* %labelreturn, align [[#SBYTES]]
   ; CHECK: [[RO:%.*]] = load i32, i32* %originreturn, align 4
   ; CHECK: store i[[#SBITS]] [[RS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_retval_tls to i[[#SBITS]]*), align 2
@@ -215,7 +215,7 @@
   ; CHECK: [[AO:%.*]] = load i32, i32* getelementptr inbounds ([200 x i32], [200 x i32]* @__dfsan_arg_origin_tls, i64 0, i64 0), align 4
   ; CHECK: [[BS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 2) to i[[#SBITS]]*), align 2
   ; CHECK: [[AS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_arg_tls to i[[#SBITS]]*), align 2
-  ; CHECK: call void @__dfso_custom_cb_without_ret(void (void (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i32, i32)* @"dfst0$custom_cb_without_ret", i8* bitcast (void (i32, i32)* @cb_without_ret.dfsan to i8*), i32 %a, i32 %b, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i32 zeroext 0, i32 zeroext [[AO]], i32 zeroext [[BO]])
+  ; CHECK: call void @__dfso_custom_cb_without_ret(void (i32, i32)* @cb_without_ret.dfsan, i32 %a, i32 %b, i[[#SBITS]] zeroext 0, i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i32 zeroext 0, i32 zeroext [[AO]], i32 zeroext [[BO]])
   ; CHECK-NEXT: ret void
 
   call void @custom_cb_without_ret(void (i32, i32)* @cb_without_ret, i32 %a, i32 %b)
@@ -267,8 +267,7 @@
 ; CHECK-NEXT:  [[BS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 4) to i[[#SBITS]]*), align 2
 ; CHECK-NEXT:  [[AS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 2) to i[[#SBITS]]*), align 2
 ; CHECK-NEXT:  [[CS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_arg_tls to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  [[C:%.*]] = bitcast i32 (i32, i32)* %0 to i8*
-; CHECK-NEXT:  [[R:%.*]] = call i32 @__dfso_custom_cb_with_ret(i32 (i32 (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32*)* @"dfst0$custom_cb_with_ret", i8* [[C]], i32 %1, i32 %2, i[[#SBITS]] zeroext [[CS]], i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i[[#SBITS]]* %labelreturn, i32 zeroext [[CO]], i32 zeroext [[AO]], i32 zeroext [[BO]], i32* %originreturn)
+; CHECK-NEXT:  [[R:%.*]] = call i32 @__dfso_custom_cb_with_ret(i32 (i32, i32)* %0, i32 %1, i32 %2, i[[#SBITS]] zeroext [[CS]], i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i[[#SBITS]]* %labelreturn, i32 zeroext [[CO]], i32 zeroext [[AO]], i32 zeroext [[BO]], i32* %originreturn)
 ; CHECK-NEXT:  [[RS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* %labelreturn, align [[#SBYTES]]
 ; CHECK-NEXT:  [[RO:%.*]] = load i32, i32* %originreturn, align 4
 ; CHECK-NEXT:  store i[[#SBITS]] [[RS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_retval_tls to i[[#SBITS]]*), align 2
@@ -282,38 +281,17 @@
 ; CHECK-NEXT:  [[BS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 4) to i[[#SBITS]]*), align 2
 ; CHECK-NEXT:  [[AS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 2) to i[[#SBITS]]*), align 2
 ; CHECK-NEXT:  [[CS:%.*]] = load i[[#SBITS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_arg_tls to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  [[C:%.*]] = bitcast void (i32, i32)* %0 to i8*
-; CHECK-NEXT:  call void @__dfso_custom_cb_without_ret(void (void (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i32, i32)* @"dfst0$custom_cb_without_ret", i8* [[C]], i32 %1, i32 %2, i[[#SBITS]] zeroext [[CS]], i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i32 zeroext [[CO]], i32 zeroext [[AO]], i32 zeroext [[BO]])
+; CHECK-NEXT:  call void @__dfso_custom_cb_without_ret(void (i32, i32)* %0, i32 %1, i32 %2, i[[#SBITS]] zeroext [[CS]], i[[#SBITS]] zeroext [[AS]], i[[#SBITS]] zeroext [[BS]], i32 zeroext [[CO]], i32 zeroext [[AO]], i32 zeroext [[BO]])
 ; CHECK-NEXT:  ret void
 
 ; CHECK: declare void @__dfso_custom_without_ret(i32, i32, i[[#SBITS]], i[[#SBITS]], i32, i32)
 
 ; CHECK: declare i32 @__dfso_custom_with_ret(i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32*)
 
-; CHECK: declare i32 @__dfso_custom_cb_with_ret(i32 (i32 (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32*)*, i8*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32, i32*)
-
-; CHECK: define linkonce_odr i32 @"dfst0$custom_cb_with_ret"(i32 (i32, i32)* %0, i32 %1, i32 %2, i[[#SBITS]] %3, i[[#SBITS]] %4, i[[#SBITS]]* %5, i32 %6, i32 %7, i32* %8)
-; CHECK:   store i32 %6, i32* getelementptr inbounds ([200 x i32], [200 x i32]* @__dfsan_arg_origin_tls, i64 0, i64 0), align 4
-; CHECK-NEXT:  store i[[#SBITS]] %3, i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_arg_tls to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  store i32 %7, i32* getelementptr inbounds ([200 x i32], [200 x i32]* @__dfsan_arg_origin_tls, i64 0, i64 1), align 4
-; CHECK-NEXT:  store i[[#SBITS]] %4, i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 2) to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  %9 = call i32 %0(i32 %1, i32 %2)
-; CHECK-NEXT:  %_dfsret = load i[[#SBITS]], i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_retval_tls to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  %_dfsret_o = load i32, i32* @__dfsan_retval_origin_tls, align 4
-; CHECK-NEXT:  store i[[#SBITS]] %_dfsret, i[[#SBITS]]* %5, align [[#SBYTES]]
-; CHECK-NEXT:  store i32 %_dfsret_o, i32* %8, align 4
-; CHECK-NEXT:  ret i32 %9
-
-; CHECK: declare void @__dfso_custom_cb_without_ret(void (void (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i32, i32)*, i8*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]], i32, i32, i32)
-
-; CHECK: define linkonce_odr void @"dfst0$custom_cb_without_ret"(void (i32, i32)* %0, i32 %1, i32 %2, i[[#SBITS]] %3, i[[#SBITS]] %4, i32 %5, i32 %6)
-; CHECK:  store i32 %5, i32* getelementptr inbounds ([200 x i32], [200 x i32]* @__dfsan_arg_origin_tls, i64 0, i64 0), align 4
-; CHECK-NEXT:  store i[[#SBITS]] %3, i[[#SBITS]]* bitcast ([[TLS_ARR]]* @__dfsan_arg_tls to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  store i32 %6, i32* getelementptr inbounds ([200 x i32], [200 x i32]* @__dfsan_arg_origin_tls, i64 0, i64 1), align 4
-; CHECK-NEXT:  store i[[#SBITS]] %4, i[[#SBITS]]* inttoptr (i64 add (i64 ptrtoint ([[TLS_ARR]]* @__dfsan_arg_tls to i64), i64 2) to i[[#SBITS]]*), align 2
-; CHECK-NEXT:  call void %0(i32 %1, i32 %2)
-; CHECK-NEXT:  ret void
+; CHECK: declare i32 @__dfso_custom_cb_with_ret(i32 (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32, i32*)
+
+; CHECK: declare void @__dfso_custom_cb_without_ret(void (i32, i32)*, i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]], i32, i32, i32)
 
 ; CHECK: declare void @__dfso_custom_varg_without_ret(i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i32, i32, i32*, ...)
 
-; CHECK: declare i32 @__dfso_custom_varg_with_ret(i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i[[#SBITS]]*, i32, i32, i32*, i32*, ...)
\ No newline at end of file
+; CHECK: declare i32 @__dfso_custom_varg_with_ret(i32, i32, i[[#SBITS]], i[[#SBITS]], i[[#SBITS]]*, i[[#SBITS]]*, i32, i32, i32*, i32*, ...)