[dfsan] Support Linux loongarch64

[Ami-zhang]

Summary

Make minor changes to enable DFSAN and its tests on
loongarch64. And port Linux loongarch64 memory mappings
from msan.

[browneee]

Who will use DFSan on loongarch64?

Who will maintain DFSan on loongarch64?

Is there a loongarch64 buildbot?

[SixWeining]

In D140690#4017693, @browneee wrote:

Who will use DFSan on loongarch64?

Who will maintain DFSan on loongarch64?

Is there a loongarch64 buildbot?

Hi @browneee, I'm the co-author of this patch. Thanks for your comments.

Who will use DFSan on loongarch64?

Currently nobody uses it as loongarch is a pretty new arch which is expected to graduate from experimental before llvm 16 release. (But I think there may be potential users of DFSan in future.) The motivation of this change is to make loongarch64 support as more compiler-rt components as possible. We notice that DFSan is a work in progress, currently under development for x86_64 Linux. So maybe we can defer the change until DFSan development is finished and there is a use case on loongarch64?

Who will maintain DFSan on loongarch64?

Loongson (https://www.loongson.cn/EN) can maintain it.

Is there a loongarch64 buildbot?

Yes, there is one: https://lab.llvm.org/staging/#/builders/236.

[browneee]

In D140690#4018043, @SixWeining wrote:

We notice that DFSan is a work in progress, currently under development for x86_64 Linux. So maybe we can defer the change until DFSan development is finished and there is a use case on loongarch64?

This sounds like a reasonable option.

DFSan development is now fairly mature. We don't support other platforms due to lack of demand (all the users I'm aware of are x86_64).

As a side note: I am pleased to see that the patch required to add this support is as small as I'd expect (even if we add the pieces in D140689 not shown here atm).

[MaskRay]

In D140690#4018617, @browneee wrote:

In D140690#4018043, @SixWeining wrote:

We notice that DFSan is a work in progress, currently under development for x86_64 Linux. So maybe we can defer the change until DFSan development is finished and there is a use case on loongarch64?

This sounds like a reasonable option.

DFSan development is now fairly mature. We don't support other platforms due to lack of demand (all the users I'm aware of are x86_64).

As a side note: I am pleased to see that the patch required to add this support is as small as I'd expect (even if we add the pieces in D140689 not shown here atm).

@browneee Instead of marking most tests as REQUIRES: x86_64-target-arch, it'd be better to use an opt-out model (UNSUPPORTED: ... or XFAIL: ...)? The majority of tests should be architecture-agnostic.

[MaskRay]

This should be rebased after D140744 removed // REQUIRES: x86_64-target-arch

[tangyouling]

The patch will continue to be updated by @Ami-zhang

[Ami-zhang]

rebase

clang/lib/Driver/ToolChains/Linux.cpp

SanitizerMask Linux::getSupportedSanitizers() const {
   Res |= SanitizerKind::PointerCompare;
   Res |= SanitizerKind::PointerSubtract;
   Res |= SanitizerKind::Fuzzer;
   Res |= SanitizerKind::FuzzerNoLink;
   Res |= SanitizerKind::KernelAddress;
   Res |= SanitizerKind::Memory;
   Res |= SanitizerKind::Vptr;
   Res |= SanitizerKind::SafeStack;
-  if (IsX86_64 || IsMIPS64 || IsAArch64)
+  if (IsX86_64 || IsMIPS64 || IsAArch64 || IsLoongArch64)
     Res |= SanitizerKind::DataFlow;
   if (IsX86_64 || IsMIPS64 || IsAArch64 || IsX86 || IsArmArch || IsPowerPC64 ||
       IsRISCV64 || IsSystemZ || IsHexagon || IsLoongArch64)
     Res |= SanitizerKind::Leak;
   if (IsX86_64 || IsMIPS64 || IsAArch64 || IsPowerPC64 || IsSystemZ ||
       IsLoongArch64)
     Res |= SanitizerKind::Thread;
   if (IsX86_64 || IsSystemZ)

compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake

 
 set(ALL_SANITIZER_COMMON_SUPPORTED_ARCH ${X86} ${X86_64} ${PPC64} ${RISCV64}
     ${ARM32} ${ARM64} ${MIPS32} ${MIPS64} ${S390X} ${SPARC} ${SPARCV9}
     ${HEXAGON} ${LOONGARCH64})
 set(ALL_ASAN_SUPPORTED_ARCH ${X86} ${X86_64} ${ARM32} ${ARM64} ${RISCV64}
     ${MIPS32} ${MIPS64} ${PPC64} ${S390X} ${SPARC} ${SPARCV9} ${HEXAGON}
     ${LOONGARCH64})
 set(ALL_ASAN_ABI_SUPPORTED_ARCH ${X86_64} ${ARM64})
-set(ALL_DFSAN_SUPPORTED_ARCH ${X86_64} ${MIPS64} ${ARM64})
+set(ALL_DFSAN_SUPPORTED_ARCH ${X86_64} ${MIPS64} ${ARM64} ${LOONGARCH64})
 
 if(ANDROID)
   set(OS_NAME "Android")
 else()
   set(OS_NAME "${CMAKE_SYSTEM_NAME}")
 endif()
 
 if(OS_NAME MATCHES "Linux")

compiler-rt/test/dfsan/lit.cfg.py

 
 config.substitutions.append(("%clang_dfsan ", build_invocation(clang_dfsan_cflags)))
 config.substitutions.append(("%clangxx_dfsan ", build_invocation(clang_dfsan_cxxflags)))
 
 # Default test suffixes.
 config.suffixes = [".c", ".cpp"]
 
 # DataFlowSanitizer tests are currently supported on Linux only.
-if not (config.host_os in ["Linux"] and config.target_arch in ["aarch64", "x86_64"]):
+if not (config.host_os in ["Linux"] and config.target_arch in ["aarch64", "x86_64", "loongarch64"]):
     config.unsupported = True

llvm/lib/Transforms/Instrumentation/DataFlowSanitizer.cpp

 const MemoryMapParams Linux_X86_64_MemoryMapParams = {
     0,              // AndMask (not used)
     0x500000000000, // XorMask
     0,              // ShadowBase (not used)
     0x100000000000, // OriginBase
 };
 // NOLINTEND(readability-identifier-naming)
 
+// loongarch64 Linux
+const MemoryMapParams Linux_LoongArch64_MemoryMapParams = {
+    0,              // AndMask (not used)
+    0x500000000000, // XorMask
+    0,              // ShadowBase (not used)
+    0x100000000000, // OriginBase
+};
+
 namespace {
 
 class DFSanABIList {
   std::unique_ptr<SpecialCaseList> SCL;
 
 public:
   DFSanABIList() = default;
 
if (TargetTriple.getOS() != Triple::Linux)
     report_fatal_error("unsupported operating system");
   switch (TargetTriple.getArch()) {
   case Triple::aarch64:
     MapParams = &Linux_AArch64_MemoryMapParams;
     break;
   case Triple::x86_64:
     MapParams = &Linux_X86_64_MemoryMapParams;
     break;
+  case Triple::loongarch64:
+    MapParams = &Linux_LoongArch64_MemoryMapParams;
+    break;
   default:
     report_fatal_error("unsupported architecture");
   }
 
   Mod = &M;
   Ctx = &M.getContext();
   Int8Ptr = Type::getInt8PtrTy(*Ctx);
   OriginTy = IntegerType::get(*Ctx, OriginWidthBits);