[WebAssembly] Error out on indirect uses of setjmp

[aheejin]

Summary

Both Wasm & Emscripten SjLj handling has a restriction that setjmp
cannot be called indirectly. I thought we have been erroring out on
indirect uses of setjmp, but some recent CL disrupted the logic and
we are not erroring out anymore.

We currently

1. Collect functions that contain setjmp calls in SetjmpUsers. This only counts direct calls: https://github.com/llvm/llvm-project/blob/8f77dc459e31aad6daab89a124fa92067916274c/llvm/lib/Target/WebAssembly/WebAssemblyLowerEmscriptenEHSjLj.cpp#L869-L878
2. Run runSjLjOnFunction only on those SetjmpUsers. Within runSjLjOnFunction, if we see an indirect use of setjmp, we error out: https://github.com/llvm/llvm-project/blob/8f77dc459e31aad6daab89a124fa92067916274c/llvm/lib/Target/WebAssembly/WebAssemblyLowerEmscriptenEHSjLj.cpp#L1218-L1221

So if there are only indirect setjmp calls within the module,
SetjmpUsers will be empty, and runSjLjOnFunction is not even entered
once. And the indirect setjmp call will error out at link time. So in
this CL we check for the indirect uses of setjmp upfront before we
enter runSjLjOnFunction.

Also this currently errors out on invoke @setjmp, which can only occur
when using Wasm EH + Wasm SjLj within a function. We recently added Wasm
SjLj support but we don't support using Wasm EH + Wasm SjLj in the same
function yet. We plan to add this support very soon, so I don't think
it's worth creating another test file just for this. (This is an error
test so it needs its own file)

[dschuff]

llvm/lib/Target/WebAssembly/WebAssemblyLowerEmscriptenEHSjLj.cpp

881

This error could probably be a bit clearer. e.g. what exactly does not support indirect use of setjmp? Maybe "wasm does not support indirect use of setjmp" or even just "indirect use of setjmp is not supported"

[aheejin]

Change the error message

[thakis]

This seems to break tests: http://45.33.8.238/linux/55329/step_12.txt

Please take a look!

[xiangzhangllvm]

This test don't pass
x64 windows > LLVM.CodeGen/WebAssembly::lower-em-sjlj-indirect-setjmp.ll

[aheejin]

Oh sorry, I forgot to change the test, and somehow missed the breakage notifications. And thank you @MaskRay for fixing this.

llvm/lib/Target/WebAssembly/WebAssemblyLowerEmscriptenEHSjLj.cpp

if (EnableEmEH) {
     FunctionType *EHTypeIDTy =
         FunctionType::get(IRB.getInt32Ty(), IRB.getInt8PtrTy(), false);
     EHTypeIDF = getEmscriptenFunction(EHTypeIDTy, "llvm_eh_typeid_for", &M);
   }
 
   if ((EnableEmSjLj || EnableWasmSjLj) && SetjmpF) {
     // Precompute setjmp users
     for (User *U : SetjmpF->users()) {
-      if (auto *UI = dyn_cast<Instruction>(U)) {
-        auto *UserF = UI->getFunction();
+      if (auto *CB = dyn_cast<CallBase>(U)) {
+        auto *UserF = CB->getFunction();
         // If a function that calls setjmp does not contain any other calls that
         // can longjmp, we don't need to do any transformation on that function,
         // so can ignore it
         if (containsLongjmpableCalls(UserF))
           SetjmpUsers.insert(UserF);
+      } else {
+        std::string S;
+        raw_string_ostream SS(S);
+        SS << *U;
+        report_fatal_error(Twine("Indirect use of setjmp is not supported: ") +
+                           SS.str());
       }
     }
   }
 
   bool SetjmpUsed = SetjmpF && !SetjmpUsers.empty();
   bool LongjmpUsed = LongjmpF && !LongjmpF->use_empty();
   DoSjLj = (EnableEmSjLj | EnableWasmSjLj) && (SetjmpUsed || LongjmpUsed);
 
bool WebAssemblyLowerEmscriptenEHSjLj::runSjLjOnFunction(Function &F) {
   SetjmpTableInsts.push_back(SetjmpTable);
   SetjmpTableSizeInsts.push_back(SetjmpTableSize);
 
   // Setjmp transformation
   SmallVector<PHINode *, 4> SetjmpRetPHIs;
   Function *SetjmpF = M.getFunction("setjmp");
   for (User *U : SetjmpF->users()) {
     auto *CI = dyn_cast<CallInst>(U);
+    // FIXME 'invoke' to setjmp can happen when we use Wasm EH + Wasm SjLj, but
+    // we don't support two being used together yet.
     if (!CI)
-      report_fatal_error("Does not support indirect calls to setjmp");
-
+      report_fatal_error("Wasm EH + Wasm SjLj is not fully supported yet");
     BasicBlock *BB = CI->getParent();
     if (BB->getParent() != &F) // in other function
       continue;
 
     // The tail is everything right after the call, and will be reached once
     // when setjmp is called, and later when longjmp returns to the setjmp
     BasicBlock *Tail = SplitBlock(BB, CI->getNextNode());
     // Add a phi to the tail, which will be the output of setjmp, which

llvm/test/CodeGen/WebAssembly/lower-em-sjlj-indirect-setjmp.ll

+; RUN: not --crash llc < %s -enable-emscripten-sjlj 2>&1 | FileCheck %s
+; RUN: not --crash llc < %s -wasm-enable-sjlj -mattr=+exception-handling -exception-model=wasm 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+%struct.__jmp_buf_tag = type { [6 x i32], i32, [32 x i32] }
+
+; CHECK: LLVM ERROR: Does not support indirect uses of setjmp
+@setjmp_fp = global i32 (%struct.__jmp_buf_tag*)* @setjmp, align 4
+
+define void @indirect_setjmp_call() {
+entry:
+  %buf = alloca [1 x %struct.__jmp_buf_tag], align 16
+  %0 = load i32 (%struct.__jmp_buf_tag*)*, i32 (%struct.__jmp_buf_tag*)** @setjmp_fp, align 4
+  %arraydecay = getelementptr inbounds [1 x %struct.__jmp_buf_tag], [1 x %struct.__jmp_buf_tag]* %buf, i32 0, i32 0
+  %call = call i32 %0(%struct.__jmp_buf_tag* %arraydecay)
+  call void @foo()
+  ret void
+}
+
+declare void @foo()
+; Function Attrs: returns_twice
+declare i32 @setjmp(%struct.__jmp_buf_tag*) #0
+
+attributes #0 = { returns_twice }
+