[LTO][Legacy] Add new API to check presence of ctor/dtor functions

[w2yehia]

Summary

On AIX, the linker needs to check whether a given lto_module_t contains any constructor/destructor functions, in order to implement the behavior of the -bcdtors:all flag. See https://www.ibm.com/docs/en/aix/7.2?topic=l-ld-command for the flag's documentation.
In llvm IR, constructor (destructor) functions are added to a special global array @llvm.global_ctors (@llvm.global_dtors).
However, because these two symbols are artificial, they are not visited during the symbol traversal (using the lto_module_get_[num_symbols|symbol_name|symbol_attribute] API).

This patch adds a new function to the libLTO interface that checks the presence of one or both of these two symbols.

[steven_wu]

Comment inline.

It would be good if you can find a way to write testcase.

llvm/lib/LTO/LTOModule.cpp

695

Can you simply this code a bit by using consume_front() on the name?

Also you need to check for nullptr for PointerUnion:

if (auto *GV = Sym.dyn_cast<>)

[w2yehia]

address code review comment.

[w2yehia]

Added testcase for LTOModule::hasCtorDtor()

@steven_wu FYI

[steven_wu]

LGTM in general. Just two small comments inline. Thanks!

llvm/tools/llvm-lto/llvm-lto.cpp

404

Can you update the comments here to reflect the current function?

420

Can you make the output to stdout more verbose? Filename : {0|1} is not clear even for a testing tool.

[w2yehia]

commited version

llvm/include/llvm-c/lto.h

  * @{
  */
 
-#define LTO_API_VERSION 28
+#define LTO_API_VERSION 29
 
 /**
  * \since prior to LTO_API_VERSION=3

                                                unsigned int *out_cputype,
                                                unsigned int *out_cpusubtype);
 
+/**
+ * This function can be used by the linker to check if a given module has
+ * any constructor or destructor functions.
+ *
+ * Returns true if the module has either the @llvm.global_ctors or the
+ * @llvm.global_dtors symbol. Otherwise returns false.
+ *
+ * \since LTO_API_VERSION=29
+ */
+extern lto_bool_t lto_module_has_ctor_dtor(lto_module_t mod);
 /**
  * Diagnostic severity.
  *

llvm/include/llvm/LTO/legacy/LTOModule.h

 
   Expected<uint32_t> getMachOCPUSubType() const;
 
+  /// Returns true if the module has either the @llvm.global_ctors or the
+  /// @llvm.global_dtors symbol. Otherwise returns false.
+  bool hasCtorDtor() const;
+
 private:
   /// Parse metadata from the module
   // FIXME: it only parses "llvm.linker.options" metadata at the moment

llvm/lib/LTO/LTOModule.cpp

 Expected<uint32_t> LTOModule::getMachOCPUSubType() const {
   return MachO::getCPUSubType(Triple(Mod->getTargetTriple()));
 }
+
+bool LTOModule::hasCtorDtor() const {
+  for (auto Sym : SymTab.symbols()) {
+    if (auto *GV = Sym.dyn_cast<GlobalValue *>()) {
+      StringRef Name = GV->getName();
+      if (Name.consume_front("llvm.global_")) {
+        if (Name.equals("ctors") || Name.equals("dtors"))
+          return true;
+      }
+    }
+  }
+  return false;
+}

llvm/test/tools/llvm-lto/ltomodule.ll

+# RUN: rm -rf %t && split-file %s %t
+
+; RUN: llvm-as < %t/hasCtor.ll > %t.bc
+; RUN: llvm-lto %t.bc -query-hasCtorDtor | FileCheck %s --check-prefixes=POSITIVE
+
+; RUN: llvm-as < %t/hasDtor.ll > %t.bc
+; RUN: llvm-lto %t.bc -query-hasCtorDtor | FileCheck %s --check-prefixes=POSITIVE
+
+; RUN: llvm-as < %t/hasBoth.ll > %t.bc
+; RUN: llvm-lto %t.bc -query-hasCtorDtor | FileCheck %s --check-prefixes=POSITIVE
+
+; RUN: llvm-as < %t/hasNone.ll > %t.bc
+; RUN: llvm-lto %t.bc -query-hasCtorDtor | FileCheck %s --check-prefixes=NEGATIVE
+
+; POSITIVE: .bc: hasCtorDtor = true
+; NEGATIVE: .bc: hasCtorDtor = false
+
+;--- hasCtor.ll
[e-mail]_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @constructor, i8* null }]
+declare void @constructor()
+
+;--- hasDtor.ll
[e-mail]_dtors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @destructor, i8* null }]
+declare void @destructor()
+
+;--- hasBoth.ll
[e-mail]_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @constructor, i8* null }]
[e-mail]_dtors = appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 65535, void ()* @destructor, i8* null }]
+declare void @constructor()
+declare void @destructor()
+
+;--- hasNone.ll
+declare void @foo()
+
+

llvm/tools/llvm-lto/llvm-lto.cpp

         "Instead of running LTO, list the dependent libraries in each IR file"),
     cl::cat(LTOCategory));
 
+static cl::opt<bool> QueryHasCtorDtor(
+    "query-hasCtorDtor", cl::init(false),
+    cl::desc("Queries LTOModule::hasCtorDtor() on each IR file"));
+
 static cl::opt<bool>
     SetMergedModule("set-merged-module", cl::init(false),
                     cl::desc("Use the first input module as the merged module"),

   }
 }
 
-/// List symbols in each IR file.
+/// Load each IR file and dump certain information based on active flags.
 ///
 /// The main point here is to provide lit-testable coverage for the LTOModule
-/// functionality that's exposed by the C API to list symbols.  Moreover, this
-/// provides testing coverage for modules that have been created in their own
-/// contexts.
-static void listSymbols(const TargetOptions &Options) {
+/// functionality that's exposed by the C API. Moreover, this provides testing
+/// coverage for modules that have been created in their own contexts.
+static void testLTOModule(const TargetOptions &Options) {
   for (auto &Filename : InputFilenames) {
     std::unique_ptr<MemoryBuffer> Buffer;
     std::unique_ptr<LTOModule> Module =
         getLocalLTOModule(Filename, Buffer, Options);
 
-    // List the symbols.
-    outs() << Filename << ":\n";
-    for (int I = 0, E = Module->getSymbolCount(); I != E; ++I)
-      outs() << Module->getSymbolName(I) << "\n";
+    if (ListSymbolsOnly) {
+      // List the symbols.
+      outs() << Filename << ":\n";
+      for (int I = 0, E = Module->getSymbolCount(); I != E; ++I)
+        outs() << Module->getSymbolName(I) << "\n";
+    }
+    if (QueryHasCtorDtor)
+      outs() << Filename
+             << ": hasCtorDtor = " << (Module->hasCtorDtor() ? "true" : "false")
+             << "\n";
   }
 }
 

   // set up the TargetOptions for the machine
   TargetOptions Options = codegen::InitTargetOptionsFromCodeGenFlags(Triple());
 
-  if (ListSymbolsOnly) {
-    listSymbols(Options);
+  if (ListSymbolsOnly || QueryHasCtorDtor) {
+    testLTOModule(Options);
     return 0;
   }
 

llvm/tools/lto/lto.cpp

   unwrap(cg)->setShouldEmbedUselists(ShouldEmbedUselists);
 }
 
+lto_bool_t lto_module_has_ctor_dtor(lto_module_t mod) {
+  return unwrap(mod)->hasCtorDtor();
+}
+
 // ThinLTO API below
 
 thinlto_code_gen_t thinlto_create_codegen(void) {

llvm/tools/lto/lto.exports

 lto_module_create_from_memory_with_path
 lto_module_create_in_local_context
 lto_module_create_in_codegen_context
+lto_module_has_ctor_dtor
 lto_module_get_linkeropts
 lto_module_get_macho_cputype
 lto_module_get_num_symbols