Refactor

src/mold.h

@@ -1551,7 +1551,6 @@ private:
 
 template <> u64 get_eflags(Context<ARM32> &ctx);
 void create_arm_exidx_section(Context<ARM32> &ctx);
-void fixup_arm_exidx_section(Context<ARM32> &ctx);
 
 //
 // arch-riscv.cc

src/passes.cc

@@ -1347,11 +1347,9 @@ void compute_section_sizes(Context<E> &ctx) {
   Timer t(ctx, "compute_section_sizes");
 
   if constexpr (needs_thunk<E>) {
-    // Chunk<E>::compute_section_size obtains a global lock to create
-    // range extension thunks. I don't know why, but using parallel_for
-    // loop both inside and outside of the lock may cause a deadlock. It
-    // might be a bug in TBB. For now, I'll avoid using parallel_for_each
-    // here.
+    // We cannot use parallel-for for compute_section_size() which may
+    // call create_range_extension_thunks() because that function is
+    // not thread-safe.
     for (Chunk<E> *chunk : ctx.chunks)
       if (chunk->shdr.sh_flags & SHF_EXECINSTR)
         chunk->compute_section_size(ctx);

src/thunks.cc

@@ -167,10 +167,6 @@ static void scan_rels(Context<E> &ctx, InputSection<E> &isec,
 
 template <>
 void OutputSection<E>::create_range_extension_thunks(Context<E> &ctx) {
-  // This function is not thread-safe because it mutates symbols' members
-  static std::mutex mu;
-  std::scoped_lock lock(mu);
-
   std::span<InputSection<E> *> m = members;
   if (m.empty())
     return;