Allow building with scudo memory allocator on Windows

[russell.gallop]

Summary

Following on from https://reviews.llvm.org/D96120, this allows scudo to be used as a memory allocator for Windows.

For example, adding:

-DLLVM_INTEGRATED_CRT_ALLOC="<folder>/llvm-project/stage1/lib/clang/13.0.0/lib/windows/clang_rt.scudo-x86_64.lib"

Note that -DLLVM_INTEGRATED_CRT_ALLOC only allows hooking in C library so this doesn't replace C++ new/delete etc.

check-all passes with scudo. Intermittently see failures in:

ExecutionEngine/MCJIT/test-global-init-nonzero-sm-pic.ll
ExecutionEngine/MCJIT/test-ptr-reloc-sm-pic.ll
ExecutionEngine/MCJIT/multi-module-sm-pic-a.ll
ExecutionEngine/MCJIT/cross-module-sm-pic-a.ll

Which appear to be due to https://bugs.llvm.org/show_bug.cgi?id=24978 (see analysis here: https://reviews.llvm.org/D86694#2539630). This bug happens more frequently with Scudo due to the difference in allocated addresses.

[russell.gallop]

Add documentation to CMake.rst

[aganea]

llvm/docs/CMake.rst

496

How did you end up fixing the new/delete link issues in https://reviews.llvm.org/D86694#2463429 ?

llvm/lib/Support/CMakeLists.txt

70

Strangely we don't #include <rpnew.h>[1] anywhere when activating rpmalloc, and that still works (I don't see calls to Win32 Heap functions in the profile traces, only calls to rpmalloc functions). I think this is because the default behavior of the Windows CRT new/delete is to call malloc/free. So I'm not quite sure in which situation we need clang_rt.scudo_cxx / if it is needed here?

[1] https://github.com/mjansson/rpmalloc/blob/develop/rpmalloc/rpnew.h

[russell.gallop]

llvm/docs/CMake.rst

496

-fsanitize adds both to the link if going via the clang driver: https://reviews.llvm.org/D96120#change-6VXGONdzQcwB

This works for the lit tests, but -fsanitize doesn't work for building LLVM on Windows as the link steps don't go via the compiler, and wouldn't work with cl.exe.

llvm/lib/Support/CMakeLists.txt

70

Yes, I don't think that they're "required" for scudo to take over memory allocations via malloc. I believe that Windows new then goes to malloc which ends up with scudoAllocate.

So I'm not quite sure in which situation we need clang_rt.scudo_cxx / if it is needed here?

AllocType mismatch detection won't work between new->free and malloc->delete.

[aganea]

llvm/lib/Support/CMakeLists.txt

70

Ah I see - would that trigger any asserts in scudo? (if scudo was compiled with -DLLVM_ENABLE_ASSERTIONS=ON). In that case, should we add the second lib below? ie.

STRING(REGEX REPLACE "(.+)\\.scudo-(.+)" "\\1.scudo_cxx-\\2" LLVM_INTEGRATED_CRT_ALLOC2 ${LLVM_INTEGRATED_CRT_ALLOC})
set(system_libs ${system_libs} "${LLVM_INTEGRATED_CRT_ALLOC2}")

[russell.gallop]

llvm/lib/Support/CMakeLists.txt

70

Ah I see - would that trigger any asserts in scudo?

I don't think assertions, but errors.

In that case, should we add the second lib below?

I tried similar to and it fails with a multiply defined symbol error:

clang_rt.scudo_cxx-x86_64.lib(scudo_new_delete.cpp.obj) : error LNK2005: "void * __cdecl operator new(unsigned __int64)" (??2@YAPEAX_K@Z) already defined in libclang.lib(libclang.dll)
clang_rt.scudo_cxx-x86_64.lib(scudo_new_delete.cpp.obj) : error LNK2005: "void * __cdecl operator new(unsigned __int64,struct std::nothrow_t const &)" (??2@YAPEAX_KAEBUnothrow_t@std@@@Z) already defined in libclang.lib(libclang.dll)
clang_rt.scudo_cxx-x86_64.lib(scudo_new_delete.cpp.obj) : error LNK2005: "void __cdecl operator delete(void *)" (??3@YAXPEAX@Z) already defined in libclang.lib(libclang.dll)
clang_rt.scudo_cxx-x86_64.lib(scudo_new_delete.cpp.obj) : error LNK2005: "void __cdecl operator delete(void *,unsigned __int64)" (??3@YAXPEAX_K@Z) already defined in libclang.lib(libclang.dll)
clang_rt.scudo_cxx-x86_64.lib(scudo_new_delete.cpp.obj) : error LNK2005: "void * __cdecl operator new[](unsigned __int64)" (??_U@YAPEAX_K@Z) already defined in libclang.lib(libclang.dll)
clang_rt.scudo_cxx-x86_64.lib(scudo_new_delete.cpp.obj) : error LNK2005: "void __cdecl operator delete[](void *)" (??_V@YAXPEAX@Z) already defined in libclang.lib(libclang.dll)
   Creating library lib\c-index-test.lib and object lib\c-index-test.exp
bin\c-index-test.exe : fatal error LNK1169: one or more multiply defined symbols found

I'm not sure whether it is integrated correctly for the rest of the executables. I can take another look.

llvm/docs/CMake.rst

**LLVM_USE_CRT_{target}**:STRING
   On Windows, tells which version of the C runtime library (CRT) should be used.
   For example, -DLLVM_USE_CRT_RELEASE=MT would statically link the CRT into the
   LLVM tools and library.
 
 **LLVM_INTEGRATED_CRT_ALLOC**:PATH
   On Windows, allows embedding a different C runtime allocator into the LLVM
   tools and libraries. Using a lock-free allocator such as the ones listed below
   greatly decreases ThinLTO link time by about an order of magnitude. It also
-  midly improves Clang build times, by about 5-10%. At the moment, rpmalloc,
-  snmalloc and mimalloc are supported. Use the path to `git clone` to select
-  the respective allocator, for example:
+  mildly improves Clang build times, by about 5-10%. At the moment, rpmalloc,
+  snmalloc and mimalloc from outside llvm-project are supported. Use the path
+  to `git clone` to select the respective allocator, for example:
 
   .. code-block:: console
 
     $ D:\git> git clone https://github.com/mjansson/rpmalloc
     $ D:\llvm-project> cmake ... -DLLVM_INTEGRATED_CRT_ALLOC=D:\git\rpmalloc
-  
+  Scudo from compiler-rt is also supported by specifying the path to the built
+  library, for example:
+
+  .. code-block:: console
+
+    $ D:\llvm-project> cmake ... -DLLVM_INTEGRATED_CRT_ALLOC=<stage1>\lib\clang\<version>\lib\windows\clang_rt.scudo-x86_64.lib
+
   This flag needs to be used along with the static CRT, ie. if building the
   Release target, add -DLLVM_USE_CRT_RELEASE=MT.
 
 **LLVM_BUILD_DOCS**:BOOL
   Adds all *enabled* documentation targets (i.e. Doxgyen and Sphinx targets) as
   dependencies of the default build targets.  This results in all of the (enabled)
   documentation targets being as part of a normal build.  If the ``install``
   target is run then this also enables all built documentation targets to be

llvm/lib/Support/CMakeLists.txt

 if(LLVM_INTEGRATED_CRT_ALLOC)
   if (CMAKE_BUILD_TYPE AND NOT ${LLVM_USE_CRT_${uppercase_CMAKE_BUILD_TYPE}} MATCHES "^(MT|MTd)$")
     message(FATAL_ERROR "LLVM_INTEGRATED_CRT_ALLOC only works with /MT or /MTd. Use LLVM_USE_CRT_${uppercase_CMAKE_BUILD_TYPE} to set the appropriate option.")
   endif()
 
   string(REGEX REPLACE "(/|\\\\)$" "" LLVM_INTEGRATED_CRT_ALLOC "${LLVM_INTEGRATED_CRT_ALLOC}")
 
   if(NOT EXISTS "${LLVM_INTEGRATED_CRT_ALLOC}")
-    message(FATAL_ERROR "Cannot find the path to `git clone` for the CRT allocator! (${LLVM_INTEGRATED_CRT_ALLOC}). Currently, rpmalloc, snmalloc and mimalloc are supported.")
+    message(FATAL_ERROR "Cannot find the path to `git clone` for the CRT allocator! (${LLVM_INTEGRATED_CRT_ALLOC}). Currently, rpmalloc, snmalloc, mimalloc and scudo are supported.")
   endif()
 
   if(LLVM_INTEGRATED_CRT_ALLOC MATCHES "rpmalloc$")
     add_definitions(-DENABLE_OVERRIDE -DENABLE_PRELOAD)
     set(ALLOCATOR_FILES "${LLVM_INTEGRATED_CRT_ALLOC}/rpmalloc/rpmalloc.c")
   elseif(LLVM_INTEGRATED_CRT_ALLOC MATCHES "snmalloc$")
     set(ALLOCATOR_FILES "${LLVM_INTEGRATED_CRT_ALLOC}/src/override/malloc.cc" "${LLVM_INTEGRATED_CRT_ALLOC}/src/override/new.cc")
     set(system_libs ${system_libs} "mincore.lib" "-INCLUDE:malloc")
   elseif(LLVM_INTEGRATED_CRT_ALLOC MATCHES "mimalloc$")
     set(MIMALLOC_LIB "${LLVM_INTEGRATED_CRT_ALLOC}/out/msvc-x64/Release/mimalloc-static.lib")
     if(NOT EXISTS "${MIMALLOC_LIB}")
     message(FATAL_ERROR "Cannot find the mimalloc static library. To build it, first apply the patch from https://github.com/microsoft/mimalloc/issues/268 then build the Release x64 target through ${LLVM_INTEGRATED_CRT_ALLOC}\\ide\\vs2019\\mimalloc.sln")
     endif()
     set(system_libs ${system_libs} "${MIMALLOC_LIB}" "-INCLUDE:malloc")
+  elseif(LLVM_INTEGRATED_CRT_ALLOC MATCHES "scudo")
+    set(system_libs ${system_libs} "${LLVM_INTEGRATED_CRT_ALLOC}" "-INCLUDE:malloc")
   endif()
 endif()
 
 add_llvm_component_library(LLVMSupport
   AArch64TargetParser.cpp
   ABIBreak.cpp
   ARMTargetParser.cpp
   AMDGPUMetadata.cpp