Guard unused llvm-libunwind symbols to avoid duplicates on Android

[sbomer]

NativeAOT uses llvm-libunwind's internal C++ classes directly
(DwarfInstructions, CompactUnwinder, UnwindCursor, LocalAddressSpace)
and does not call the public unw_* C API. These symbols conflict with
platform libunwind on Android NDK r29.

Add a _LIBUNWIND_NATIVEAOT define that guards out the public API
symbols NativeAOT does not use:

• libunwind.cpp: all unw_* functions, unw_local_addr_space, and
  the LocalAddressSpace::sThisAddressSpace singleton
• AddressSpace.hpp: sThisAddressSpace static member declaration
• Unwind-EHABI.cpp: the C++ exception dispatch functions are also
  guarded because they depend on the now-guarded unw_* functions
  and would cause undefined symbol errors on ARM32

Switch UnixNativeCodeManager to use a new UnwindHelpers::FindUnwindSections()
accessor instead of the now-guarded sThisAddressSpace singleton. This
delegates to the existing file-local LocalAddressSpace instance in
UnwindHelpers.cpp

Fixes #121172

[dotnet-policy-service]

Tagging subscribers to this area: @agocke, @dotnet/ilc-contrib
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

This PR updates the NativeAOT build of the vendored llvm-libunwind to avoid duplicate symbol conflicts (notably on Android NDK r29) by compiling out libunwind’s public unw_* entrypoints and related exception-dispatch code that NativeAOT doesn’t use, and by routing NativeAOT’s unwind-section discovery through a runtime-owned helper instead of libunwind’s LocalAddressSpace::sThisAddressSpace singleton.

Changes:

• Add _LIBUNWIND_NATIVEAOT and use it to guard out unw_* APIs (and the LocalAddressSpace::sThisAddressSpace singleton) in libunwind.cpp, plus dependent EHABI exception-dispatch exports in Unwind-EHABI.cpp.
• Introduce UnwindHelpers::FindUnwindSections() and switch UnixNativeCodeManager to use it (avoiding the now-guarded sThisAddressSpace).
• Record the patch application in llvm-libunwind-version.txt and enable the new define in the NativeAOT runtime CMake configuration.

Show a summary per file

File	Description
src/native/external/llvm-libunwind/src/Unwind-EHABI.cpp	Guards EHABI exception-dispatch exports when _LIBUNWIND_NATIVEAOT is set to avoid references to removed __unw_* APIs.
src/native/external/llvm-libunwind/src/libunwind.cpp	Guards the public unw_* API surface and the LocalAddressSpace::sThisAddressSpace singleton under _LIBUNWIND_NATIVEAOT.
src/native/external/llvm-libunwind/src/AddressSpace.hpp	Guards the LocalAddressSpace::sThisAddressSpace static member declaration under _LIBUNWIND_NATIVEAOT.
src/native/external/llvm-libunwind-version.txt	Notes the applied commit associated with this local patch.
src/coreclr/nativeaot/Runtime/unix/UnwindHelpers.h	Declares the new UnwindHelpers::FindUnwindSections accessor.
src/coreclr/nativeaot/Runtime/unix/UnwindHelpers.cpp	Implements FindUnwindSections by delegating to the existing file-scope _addressSpace.
src/coreclr/nativeaot/Runtime/unix/UnixNativeCodeManager.cpp	Replaces use of LocalAddressSpace::sThisAddressSpace.findUnwindSections with UnwindHelpers::FindUnwindSections.
src/coreclr/nativeaot/Runtime/CMakeLists.txt	Defines _LIBUNWIND_NATIVEAOT=1 for Unix NativeAOT builds to enable the guarding behavior.

Copilot's findings

• Files reviewed: 8/8 changed files
• Comments generated: 0

[am11]

If it's a negative if !defined(_LIBUNWIND_NATIVEAOT) check everywhere, we can just delete the code it skips. The idea of this local copy is we apply the patches and keep track of them for possible upstream.

[jkotas]

we can just delete the code it skips. The idea of this local copy is we apply the patches and keep track of them for possible upstream.

Deleting the code creates more conflicts during updates. We won't be able to upstream the deletion either. I think ifdefing is better.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[sbomer]

@jkotas can you please help merge this without squashing?

[MichalStrehovsky]

This broke uwinding on arm32. The failure mode is a rather mysterious sigsegv, but pulling down a binary from a failed test run and running nm -D seems to indicate we're linking dynamically against some of GCC's libunwind (@GCC_3.5):

007ca06c D DotNetRuntimeDebugHeader@@V1.0
         w _ITM_deregisterTMCloneTable
         w _ITM_registerTMCloneTable
         U _Unwind_VRS_Get@GCC_3.5
         U _Unwind_VRS_Pop@GCC_3.5
         U _Unwind_VRS_Set@GCC_3.5
         U __aeabi_unwind_cpp_pr0@GCC_3.5
         U __aeabi_unwind_cpp_pr1@GCC_3.5
         U __aeabi_unwind_cpp_pr2@GCC_3.5
         U __clock_gettime64@GLIBC_2.34
         U __clock_nanosleep_time64@GLIBC_2.34
         U __cxa_atexit@GLIBC_2.4
         w __cxa_finalize@GLIBC_2.4
         U __errno_location@GLIBC_2.4
         U __fcntl_time64@GLIBC_2.34
         U __fstat64_time64@GLIBC_2.34
         U __getdelim@GLIBC_2.4
         ... redacted for brevity

Copilot is trying to do something at #128830.