[libc++] Add CMake option to enable execute-only code generation on AArch64

[Il-Capitano]

For a full toolchain supporting execute-only code generation the runtime libraries also need to be pre-compiled with it enabled. For libc++ this can now be enabled with the LIBCXX_EXECUTE_ONLY_CODE CMake option during build configuration.

Related RFC: https://discourse.llvm.org/t/rfc-execute-only-code-support-for-runtime-libraries-on-aarch64/86180

[llvmbot]

@llvm/pr-subscribers-libcxx

Author: Csanád Hajdú (Il-Capitano)

Changes

For a full toolchain supporting execute-only code generation the runtime libraries also need to be pre-compiled with it enabled. For libc++ this can now be enabled with the LIBCXX_EXECUTE_ONLY_CODE CMake option during build configuration.

Related RFC: https://discourse.llvm.org/t/rfc-execute-only-code-support-for-runtime-libraries-on-aarch64/86180

---

Full diff: https://github.com/llvm/llvm-project/pull/140552.diff

1 Files Affected:

• (modified) libcxx/CMakeLists.txt (+13)

diff --git a/libcxx/CMakeLists.txt b/libcxx/CMakeLists.txt
index dffdd7a3c70a6..a3e0d3e909255 100644
--- a/libcxx/CMakeLists.txt
+++ b/libcxx/CMakeLists.txt
@@ -318,6 +318,8 @@ endif()
 option(LIBCXX_HERMETIC_STATIC_LIBRARY
   "Do not export any symbols from the static library." ${LIBCXX_HERMETIC_STATIC_LIBRARY_DEFAULT})
 
+option(LIBCXX_EXECUTE_ONLY_CODE "Compile libc++ as execute-only." OFF)
+
 #===============================================================================
 # Check option configurations
 #===============================================================================
@@ -538,6 +540,17 @@ function(cxx_add_basic_build_flags target)
   # errors.
   target_compile_definitions(${target} PRIVATE -D_LIBCPP_REMOVE_TRANSITIVE_INCLUDES)
 
+  if (LIBCXX_EXECUTE_ONLY_CODE)
+    target_add_compile_flags_if_supported(${target} PRIVATE -mexecute-only)
+    if (NOT CXX_SUPPORTS_MEXECUTE_ONLY_FLAG)
+      target_add_compile_flags_if_supported(${target} PRIVATE -mpure-code)
+      if (NOT CXX_SUPPORTS_MPURE_CODE_FLAG)
+        message(SEND_ERROR "Compiler doesn't support -mexecute-only or "
+                           "-mpure-code option for target ${target}!")
+      endif()
+    endif()
+  endif()
+
   if (C_SUPPORTS_COMMENT_LIB_PRAGMA)
     if (LIBCXX_HAS_PTHREAD_LIB)
       target_compile_definitions(${target} PRIVATE -D_LIBCPP_LINK_PTHREAD_LIB)

[philnik777]

Is there a reason to not do this unconditionally?

[Il-Capitano]

Is there a reason to not do this unconditionally?

Execute-only code generation for AArch64 is fairly new, and currently planned to have limited use. Also older tools may not correctly handle merging the SHF_AARCH64_PURECODE ELF section flag that LLVM puts on .text (LLD 20 and before doesn't for example. It produces a valid non-execute-only output though, just that the section flag may be incorrectly kept).

If you don't use a very recent version of Clang, LLD, and a compatible libc implementation, you won't be able to take advantage of this feature, and may get unexpected behaviour from tools. In general, this configuration is intended for assembling a complete toolchain, where you control everything.