[ARM] __cxa_end_cleanup should be called instead of _UnwindResume.

[danielkiss]

Summary

ARM EHABI[1] specifies the cxa_end_cleanup to be called after cleanup.
It will call the UnwindResume.
cxa_begin_cleanup will be called from libcxxabi while cxa_end_cleanup is never called.
This will trigger a termination when a foreign exception is processed while UnwindResume is called
because the global state will be wrong due to the missing cxa_end_cleanup call.

Additional test here: D109856
[1] https://github.com/ARM-software/abi-aa/blob/main/ehabi32/ehabi32.rst#941compiler-helper-functions

[logan]

libcxxabi/src/cxa_exception.cpp

346

s/origian/original/

llvm/lib/CodeGen/DwarfEHPrepare.cpp

233

I think in addition to isTargetEHABICompatible check, you should also check (Pers == EHPersonality::GNU_CXX || Pers == EhPersonality::GNU_CXX_SjLj) because only __gxx_personality_v0 and __gxx_personality_sj0 will call __cxa_begin_cleanup and push the exception object into the global stack.

[logan]

libcxxabi/src/cxa_exception.cpp

345

Add and its frame information after identify the caller.

llvm/test/CodeGen/ARM/eh-resume.ll

29

Please add another test for __gcc_personality_v0, which should not use __cxa_end_cleanup.

[danielkiss]

llvm/lib/CodeGen/DwarfEHPrepare.cpp

233

Right, thanks I forget about it.

[logan]

llvm/test/CodeGen/ARM/eh-resume2.ll

26

Add some comment to explain the difference between this test case and the one in eh-resume.ll.

[logan]

libcxxabi/src/cxa_exception.cpp

345

BTW, please add a period to end the sentence after "... its frame information".

Also, please re-flow the text to fit in 80 columns.

[logan]

llvm/lib/CodeGen/DwarfEHPrepare.cpp

82

Maybe it makes more sense to define this function as a member function of the Triple class?

[logan]

llvm/lib/CodeGen/DwarfEHPrepare.cpp

239

__cxa_end_cleanup does not expect the exception object. Maybe we can exploit that to eliminate unnecessary instructions that set up r0?

[danielkiss]

Moved isTargetEHABICompatible up to the Triple class.
as __cxa_end_cleanup does not take the exception object ( managed it internally ) so less code to be generated.

[logan]

llvm/lib/CodeGen/DwarfEHPrepare.cpp

82

I think renaming this as doesRewindFunctionNeedExceptionObject would be better (since it is less tied to _Unwind_Resume) and change the implementation as:

/// Determines whether the exception object must be passed to the rewind
/// function.  For example, `_Unwind_Resume` expects an exception object but
/// `__cxa_end_cleanup` does not.
bool doesRewindFunctionNeedExceptionObject() const {
  auto *FTy = RewindFunction.getFunctionType();
  return FTy->getNumParams() > 0;
}

[logan]

We are quite close now. Here are my final comments.

llvm/lib/CodeGen/DwarfEHPrepare.cpp

237

Add:

RewindFunctionCallingConv = TLI.getLibcallCallingConv(RTLIB::CXA_END_CLEANUP);

240

Add:

RewindFunctionCallingConv = TLI.getLibcallCallingConv(RTLIB::UNWIND_RESUME);

248–253

Refine as:

llvm::SmallVector<Value *, 1> RewindFunctionArgs;

if (doesRewindFunctionNeedExceptionObject())
  RewindFunctionArgs.push_back(GetExceptionObject(RI));

// Call the rewind function.
CallInst *CI = CallInst::Create(RewindFunction, RewindFunctionArgs, "",
                                UnwindBB);
CI->setCallingConv(RewindFunctionCallingConv);

(Rationale: Avoid repeating CallInst::Create(...) and CI->setCallingConv(...) for two cases.)

260–261

Move these two lines into if (doesRewindFunctionNeedExceptionObject()) (Line 271).

Add:

llvm::SmallVector<Value *, 1> RewindFunctionArgs;

279

Add:

RewindFunctionArgs.push_back(PN);

280–282

Move these out of the if body:

// Call the function.
CallInst *CI = CallInst::Create(RewindFunction, RewindFunctionArgs, "",
                                UnwindBB);
CI->setCallingConv(RewindFunctionCallingConv);

[MaskRay]

llvm/include/llvm/ADT/Triple.h

735 ↗

(On Diff #381951)

Can Darwin or Windows use EHABI related environment? If not, remove !isOSDarwin() && !isOSWindows().
If possible, check isOSBinFormatELF() instead

[mstorsjo]

llvm/include/llvm/ADT/Triple.h

735 ↗

(On Diff #381951)

At least windows can’t, afaik (I haven’t heard about anybody setting that up, and neither llvm nor libunwind support it).

[danielkiss]

llvm/lib/CodeGen/DwarfEHPrepare.cpp

237

Moved after the if (!RewindFunction) { because it need to be done if the RewindFunction is already set.
RewindFunctionCallingConv may not worth a class member..

248–253

Thanks for the rational, refactored in the spirit of it.
GetExceptionObject has sideeffects so I kept the original code structure.

[logan]

Looks good to me. Please also fix the pre-merge lint checks.

[danielkiss]

In D111703#3087326, @logan wrote:

Looks good to me. Please also fix the pre-merge lint checks.

Thanks for the reviews! Reformatted with newer clang-format.

[logan]

llvm/lib/CodeGen/DwarfEHPrepare.cpp

228–230

Thinking this further, I think we can no longer cache RewindFunction as a data member, since there may be functions with different personality functions in the same module. If we cache it in this way, we may get errors when some functions expect _Unwind_Resume and the others expect __cxa_end_cleanup (in the same module).

[danielkiss]

no longer cache RewindFunction.

[danielkiss]

[logan]

Looks good to me.

[DavidSpickett]

I think this has caused a failure on our v7 bot: https://lab.llvm.org/buildbot/#/builders/186/builds/2047

llvm/test/CodeGen/XCore/exception.ll (doesn't show up on the armv7 quick bot because it doesn't build the XCore target)

[chill]

libcxxabi/src/cxa_exception.cpp

382–384

Could/should be

pop {r1, r2, r3, r4}
mov lr, r4

then it'll be compatible with Armv6-M and Armv8.0-M.base.

[danielkiss]

@DavidSpickett, @chill Thanks, both fixed, patch relanded.
march is not enough to set Triplet.Environment and it falls back to the default, which is based on the host.

[thakis]

This breaks tests: http://45.33.8.238/linux/59368/step_12.txt

Please take a look and revert for now if it takes a while to fix.

[thakis]

Independent of the test failure: We (chromium) update clang and libcxxabi independently of each other. That works fine with this change, yes?

[logan]

Independent of the test failure: We (chromium) update clang and libcxxabi independently of each other. That works fine with this change, yes?

The libcxxabi must be integrated before the clang integration (because the __cxa_end_cleanup trampoline doesn't have unwind frame information).

[vitalybuka]

https://lab.llvm.org/buildbot/#/builders/77/builds/10938 is broken after the patch
STEP: run lit tests [arm/aosp_coral-userdebug/AOSP.MASTER] (it's green but this is UI issue)

Only this test, we found offender for the rest:

FAIL: AddressSanitizer-arm-android :: TestCases/intercept-rethrow-exception.cpp (372 of 1466)
******************** TEST 'AddressSanitizer-arm-android :: TestCases/intercept-rethrow-exception.cpp' FAILED ********************
Script:
--
: 'RUN: at line 6';     /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm-project/compiler-rt/test/sanitizer_common/android_commands/android_compile.py  /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm_build64/bin/clang  --driver-mode=g++ -fsanitize=address -mno-omit-leaf-frame-pointer -fno-omit-frame-pointer -fno-optimize-sibling-calls -gline-tables-only  --target=armv7-linux-androideabi24 --sysroot=/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/android_ndk/toolchains/llvm/prebuilt/linux-x86_64/sysroot --gcc-toolchain=/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/android_ndk/toolchains/llvm/prebuilt/linux-x86_64  -B/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/android_ndk/toolchains/llvm/prebuilt/linux-x86_64 -Wthread-safety -Wthread-safety-reference -Wthread-safety-beta -stdlib=libc++ -fuse-ld=lld  -shared-libasan -fexceptions -O0 /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm-project/compiler-rt/test/asan/TestCases/intercept-rethrow-exception.cpp -o /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/compiler_rt_build_android_arm/test/asan/ARMAndroidConfig/TestCases/Output/intercept-rethrow-exception.cpp.tmp
: 'RUN: at line 7';    /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/compiler_rt_build_android_arm/test/asan/ARMAndroidConfig/TestCases/Output/intercept-rethrow-exception.cpp.tmp
--
Exit Code: 134
Command Output (stderr):
--
+ : 'RUN: at line 6'
+ /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm-project/compiler-rt/test/sanitizer_common/android_commands/android_compile.py /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm_build64/bin/clang --driver-mode=g++ -fsanitize=address -mno-omit-leaf-frame-pointer -fno-omit-frame-pointer -fno-optimize-sibling-calls -gline-tables-only --target=armv7-linux-androideabi24 --sysroot=/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/android_ndk/toolchains/llvm/prebuilt/linux-x86_64/sysroot --gcc-toolchain=/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/android_ndk/toolchains/llvm/prebuilt/linux-x86_64 -B/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/android_ndk/toolchains/llvm/prebuilt/linux-x86_64 -Wthread-safety -Wthread-safety-reference -Wthread-safety-beta -stdlib=libc++ -fuse-ld=lld -shared-libasan -fexceptions -O0 /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/llvm-project/compiler-rt/test/asan/TestCases/intercept-rethrow-exception.cpp -o /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/compiler_rt_build_android_arm/test/asan/ARMAndroidConfig/TestCases/Output/intercept-rethrow-exception.cpp.tmp
+ : 'RUN: at line 7'
+ /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/compiler_rt_build_android_arm/test/asan/ARMAndroidConfig/TestCases/Output/intercept-rethrow-exception.cpp.tmp
Aborted 
/var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/compiler_rt_build_android_arm/test/asan/ARMAndroidConfig/TestCases/Output/intercept-rethrow-exception.cpp.script: line 2: 22341 Aborted                 (core dumped) /var/lib/buildbot/sanitizer-buildbot6/sanitizer-x86_64-linux-android/build/compiler_rt_build_android_arm/test/asan/ARMAndroidConfig/TestCases/Output/intercept-rethrow-exception.cpp.tmp
--
********************

[danielkiss]

@vitalybuka Thanks for the report. The problem could be that the libcxxabi is taken from the sysroot which contains the old code. so the compiler and the runtime is out of sync. what do think?

[vitalybuka]

In D111703#3106464, @danielkiss wrote:

@vitalybuka Thanks for the report. The problem could be that the libcxxabi is taken from the sysroot which contains the old code. so the compiler and the runtime is out of sync. what do think?

@eugenis That's very likely. We test Android with a version of NDK against compiler HEAD. I will disabled a test on android unless someone have a better idea?

[rprichard]

libcxxabi/src/cxa_exception.cpp

382–384

@danielkiss @chill

The comment above __cxa_end_cleanup says:

According to ARM EHABI 8.4.1, __cxa_end_cleanup() should not clobber any
register, thus we have to write this function in assembly so that we can save
{r1, r2, r3}.

We're currently saving/restoring r1-r3 (volatiles), but with the Armv6-M fix, we're clobbering r4 (a local/callee-save register) in order to restore lr. That seems like wrong behavior?

i.e. __cxa_end_cleanup should ensure r4 has the original value when it calls _Unwind_Resume. Before this LLVM patch, it wasn't necessary to preserve r4, because __cxa_end_cleanup_impl would do so automatically.

[danielkiss]

libcxxabi/src/cxa_exception.cpp

382–384

clobbering r4 is not in line with the aapcs. so it could be like this (not tested yet):

push	{r1, r2, r3, lr}
push	{r4}
bl	__cxa_end_cleanup_impl
pop {r1, r2, r3, r4}
mov lr, r4
pop {r4}

[chill]

libcxxabi/src/cxa_exception.cpp

382–384

Maybe need to preserve stack alignment?

push	{r4}
push	{r1, r2, r3, lr}
sub sp, #12
bl	__cxa_end_cleanup_impl
add sp, #12
pop {r1, r2, r3, r4}
mov lr, r4
pop {r4}

[rprichard]

libcxxabi/src/cxa_exception.cpp

382–384

Is preserving 16-byte alignment at every instruction boundary important, e.g. for signals/interrupts?

Also, for LIBCXXABI_BAREMETAL, we have to clobber something for the address of _Unwind_Resume. @chill had suggested ip/r12 on https://reviews.llvm.org/D113181#3108664.

Maybe this would work (also untested):

	push	{r1, r2, r3, r4}
	mov	r4, lr
	bl	__cxa_end_cleanup_impl
	mov	lr, r4
#if defined(LIBCXXABI_BAREMETAL)
	ldr	r4,	=_Unwind_Resume
	mov	ip,	r4
#endif
	pop	{r1, r2, r3, r4}
#if defined(LIBCXXABI_BAREMETAL)
	bx	ip
#else
	b	_Unwind_Resume
#endif

[rprichard]

libcxxabi/src/cxa_exception.cpp

382–384

Is preserving 16-byte alignment at every instruction boundary important, e.g. for signals/interrupts?

From AAPCS, it looks like SP must always be aligned to 4 bytes, but must be 8-byte aligned at the start of a function ("public interface"). (grep https://developer.arm.com/documentation/ihi0042 for "Universal stack constraints" and "Stack constraints at a public interface".)

[danielkiss]

libcxxabi/src/cxa_exception.cpp

382–384

Is preserving 16-byte alignment at every instruction boundary important, e.g. for signals/interrupts?

Also, for LIBCXXABI_BAREMETAL, we have to clobber something for the address of _Unwind_Resume. @chill had suggested ip/r12 on https://reviews.llvm.org/D113181#3108664.

Maybe this would work (also untested):

	push	{r1, r2, r3, r4}
	mov	r4, lr
	bl	__cxa_end_cleanup_impl
	mov	lr, r4
#if defined(LIBCXXABI_BAREMETAL)
	ldr	r4,	=_Unwind_Resume
	mov	ip,	r4
#endif
	pop	{r1, r2, r3, r4}
#if defined(LIBCXXABI_BAREMETAL)
	bx	ip
#else
	b	_Unwind_Resume
#endif

This looks good to me.

libcxxabi/src/cxa_exception.cpp

 /*
 The routine to be called after the cleanup has been performed.  It will get the
 propagating __cxa_exception from __cxa_eh_globals, and continue the stack
 unwinding with _Unwind_Resume.
 
 According to ARM EHABI 8.4.1, __cxa_end_cleanup() should not clobber any
 register, thus we have to write this function in assembly so that we can save
 {r1, r2, r3}.  We don't have to save r0 because it is the return value and the
-first argument to _Unwind_Resume().  In addition, we are saving r4 in order to
-align the stack to 16 bytes, even though it is a callee-save register.
+first argument to _Unwind_Resume().  In addition, we are saving lr in order to
+align the stack to 16 bytes and lr will be used to identify the caller and its frame information
+_Unwind_Resume never return and we need to keep the original lr so just branch to it.
 */
 __attribute__((used)) static _Unwind_Exception *
 __cxa_end_cleanup_impl()
 {
     __cxa_eh_globals* globals = __cxa_get_globals();
     __cxa_exception* exception_header = globals->propagatingExceptions;
     if (NULL == exception_header)
     {
__cxa_end_cleanup_impl()
     }
     else
     {
         globals->propagatingExceptions = NULL;
     }
     return &exception_header->unwindHeader;
 }
 
-asm (
-    "  .pushsection  .text.__cxa_end_cleanup,\"ax\",%progbits\n"
+asm("  .pushsection  .text.__cxa_end_cleanup,\"ax\",%progbits\n"
     "  .globl  __cxa_end_cleanup\n"
     "  .type __cxa_end_cleanup,%function\n"
     "__cxa_end_cleanup:\n"
-    "  push  {r1, r2, r3, r4}\n"
+    "  push  {r1, r2, r3, lr}\n"
     "  bl  __cxa_end_cleanup_impl\n"
-    "  pop {r1, r2, r3, r4}\n"
-    "  bl  _Unwind_Resume\n"
-    "  bl  abort\n"
-    "  .popsection"
-);
+    "  pop {r1, r2, r3, lr}\n"
+    "  b _Unwind_Resume\n"
+    "  .popsection");
 #endif // defined(_LIBCXXABI_ARM_EHABI)
 
 /*
 This routine can catch foreign or native exceptions.  If native, the exception
 can be a primary or dependent variety.  This routine may remain blissfully
 ignorant of whether the native exception is primary or dependent.
 
 If the exception is native:

llvm/lib/CodeGen/DwarfEHPrepare.cpp

 //===- DwarfEHPrepare - Prepare exception handling for code generation ----===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This pass mulches exception handling code into a form adapted to code
 // generation. Required if using dwarf exception handling.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/Triple.h"
 #include "llvm/Analysis/CFG.h"
 #include "llvm/Analysis/DomTreeUpdater.h"
 #include "llvm/Analysis/EHPersonalities.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/CodeGen/RuntimeLibcalls.h"
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
class DwarfEHPrepare {
 
   // RewindFunction - _Unwind_Resume or the target equivalent.
   FunctionCallee &RewindFunction;
 
   Function &F;
   const TargetLowering &TLI;
   DomTreeUpdater *DTU;
   const TargetTransformInfo *TTI;
+  const Triple &TargetTriple;
 
   /// Return the exception object from the value passed into
   /// the 'resume' instruction (typically an aggregate). Clean up any dead
   /// instructions, including the 'resume' instruction.
   Value *GetExceptionObject(ResumeInst *RI);
 
   /// Replace resumes that are not reachable from a cleanup landing pad with
   /// unreachable and then simplify those blocks.
   size_t
   pruneUnreachableResumes(SmallVectorImpl<ResumeInst *> &Resumes,
                           SmallVectorImpl<LandingPadInst *> &CleanupLPads);
 
   /// Convert the ResumeInsts that are still present
   /// into calls to the appropriate _Unwind_Resume function.
   bool InsertUnwindResumeCalls();
 
+  bool isTargetEHABICompatible(Triple TargetTriple) const {
+    return (TargetTriple.getEnvironment() == Triple::EABI ||
+            TargetTriple.getEnvironment() == Triple::GNUEABI ||
+            TargetTriple.getEnvironment() == Triple::MuslEABI ||
+            TargetTriple.getEnvironment() == Triple::EABIHF ||
+            TargetTriple.getEnvironment() == Triple::GNUEABIHF ||
+            TargetTriple.getEnvironment() == Triple::MuslEABIHF ||
+            TargetTriple.isAndroid()) &&
+           !TargetTriple.isOSDarwin() && !TargetTriple.isOSWindows();
+  }
+
 public:
   DwarfEHPrepare(CodeGenOpt::Level OptLevel_, FunctionCallee &RewindFunction_,
                  Function &F_, const TargetLowering &TLI_, DomTreeUpdater *DTU_,
-                 const TargetTransformInfo *TTI_)
+                 const TargetTransformInfo *TTI_, const Triple &TargetTriple_)
       : OptLevel(OptLevel_), RewindFunction(RewindFunction_), F(F_), TLI(TLI_),
-        DTU(DTU_), TTI(TTI_) {}
+        DTU(DTU_), TTI(TTI_), TargetTriple(TargetTriple_) {}
 
   bool run();
 };
 
 } // namespace
 
 Value *DwarfEHPrepare::GetExceptionObject(ResumeInst *RI) {
   Value *V = RI->getOperand(0);
#if LLVM_ENABLE_STATS
     NumCleanupLandingPadsRemaining -= CleanupLPads.size() - NumRemainingLPs;
 #endif
   }
 
   if (ResumesLeft == 0)
     return true; // We pruned them all.
 
   // Find the rewind function if we didn't already.
   if (!RewindFunction) {
     FunctionType *FTy =
         FunctionType::get(Type::getVoidTy(Ctx), Type::getInt8PtrTy(Ctx), false);
-    const char *RewindName = TLI.getLibcallName(RTLIB::UNWIND_RESUME);
+    const char *RewindName;
+
+    if ((Pers == EHPersonality::GNU_CXX ||
+         Pers == EHPersonality::GNU_CXX_SjLj) &&
+        isTargetEHABICompatible(TargetTriple))
+      RewindName = "__cxa_end_cleanup";
+    else
+      RewindName = TLI.getLibcallName(RTLIB::UNWIND_RESUME);
     RewindFunction = F.getParent()->getOrInsertFunction(RewindName, FTy);
   }
 
   // Create the basic block where the _Unwind_Resume call will live.
   if (ResumesLeft == 1) {
     // Instead of creating a new BB and PHI node, just append the call to
     // _Unwind_Resume to the end of the single resume block.
     ResumeInst *RI = Resumes.front();
     BasicBlock *UnwindBB = RI->getParent();
     Value *ExnObj = GetExceptionObject(RI);
 
     // Call the _Unwind_Resume function.
     CallInst *CI = CallInst::Create(RewindFunction, ExnObj, "", UnwindBB);
     CI->setCallingConv(TLI.getLibcallCallingConv(RTLIB::UNWIND_RESUME));
 
     // We never expect _Unwind_Resume to return.
     CI->setDoesNotReturn();
     new UnreachableInst(Ctx, UnwindBB);
     return true;
   }
 
   std::vector<DominatorTree::UpdateType> Updates;
   Updates.reserve(Resumes.size());
 
   BasicBlock *UnwindBB = BasicBlock::Create(Ctx, "unwind_resume", &F);
   PHINode *PN = PHINode::Create(Type::getInt8PtrTy(Ctx), ResumesLeft, "exn.obj",
                                 UnwindBB);
 
   // Extract the exception object from the ResumeInst and add it to the PHI node
   // that feeds the _Unwind_Resume call.
   for (ResumeInst *RI : Resumes) {
     BasicBlock *Parent = RI->getParent();
     BranchInst::Create(UnwindBB, Parent);
     Updates.push_back({DominatorTree::Insert, Parent, UnwindBB});
 
     Value *ExnObj = GetExceptionObject(RI);
     PN->addIncoming(ExnObj, Parent);
 
     ++NumResumesLowered;
   }
 
   // Call the function.
   CallInst *CI = CallInst::Create(RewindFunction, PN, "", UnwindBB);
   CI->setCallingConv(TLI.getLibcallCallingConv(RTLIB::UNWIND_RESUME));
 
   // We never expect _Unwind_Resume to return.
   CI->setDoesNotReturn();
   new UnreachableInst(Ctx, UnwindBB);
 
   if (DTU)
     DTU->applyUpdates(Updates);
 
   return true;
 }
 
 bool DwarfEHPrepare::run() {
   bool Changed = InsertUnwindResumeCalls();
 
   return Changed;
 }
 
 static bool prepareDwarfEH(CodeGenOpt::Level OptLevel,
                            FunctionCallee &RewindFunction, Function &F,
                            const TargetLowering &TLI, DominatorTree *DT,
-                           const TargetTransformInfo *TTI) {
+                           const TargetTransformInfo *TTI,
+                           const Triple &TargetTriple) {
   DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Lazy);
 
   return DwarfEHPrepare(OptLevel, RewindFunction, F, TLI, DT ? &DTU : nullptr,
-                        TTI)
+                        TTI, TargetTriple)
       .run();
 }
 
 namespace {
 
 class DwarfEHPrepareLegacyPass : public FunctionPass {
   // RewindFunction - _Unwind_Resume or the target equivalent.
   FunctionCallee RewindFunction = nullptr;
bool runOnFunction(Function &F) override {
     const TargetTransformInfo *TTI = nullptr;
     if (auto *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>())
       DT = &DTWP->getDomTree();
     if (OptLevel != CodeGenOpt::None) {
       if (!DT)
         DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
       TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
     }
-    return prepareDwarfEH(OptLevel, RewindFunction, F, TLI, DT, TTI);
+    return prepareDwarfEH(OptLevel, RewindFunction, F, TLI, DT, TTI,
+                          TM.getTargetTriple());
   }
 
   bool doFinalization(Module &M) override {
     RewindFunction = nullptr;
     return false;
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {

llvm/test/CodeGen/ARM/eh-resume.ll

 ; RUN: llc < %s -mtriple=armv7-apple-ios -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=IOS
 ; RUN: llc < %s -mtriple=armv7-apple-watchos -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=IOS
 ; RUN: llc < %s -mtriple=armv7k-apple-ios -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=WATCHABI
 ; RUN: llc < %s -mtriple=armv7k-apple-watchos -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=WATCHABI
+; RUN: llc < %s -mtriple=armv7-none-gnueabihf -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=EABI
+; RUN: llc < %s -mtriple=armv7-none-none -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=ABI
 
 declare void @func()
 
 declare i32 @__gxx_personality_sj0(...)
 
 define void @test0() personality i8* bitcast (i32 (...)* @__gxx_personality_sj0 to i8*) {
 entry:
   invoke void @func()
     to label %cont unwind label %lpad
 
 cont:
   ret void
 
 lpad:
   %exn = landingpad { i8*, i32 }
            cleanup
   resume { i8*, i32 } %exn
 }
 
 ; IOS: __Unwind_SjLj_Resume
 ; WATCHABI: __Unwind_Resume
+; EABI: __cxa_end_cleanup
+; ABI: _Unwind_Resume

llvm/test/CodeGen/ARM/eh-resume2.ll

+; RUN: llc < %s -mtriple=armv7-apple-ios -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=IOS
+; RUN: llc < %s -mtriple=armv7-apple-watchos -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=IOS
+; RUN: llc < %s -mtriple=armv7k-apple-ios -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=WATCHABI
+; RUN: llc < %s -mtriple=armv7k-apple-watchos -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=WATCHABI
+; RUN: llc < %s -mtriple=armv7-none-gnueabihf -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=EABI
+; RUN: llc < %s -mtriple=armv7-none-none -arm-atomic-cfg-tidy=0 | FileCheck %s -check-prefix=ABI
+
+; ARM EABI for C++/__gxx_personality* specifies __cxa_end_cleanup, but for C code / __gcc_personality*
+; the _Unwind_Resume is required.
+
+declare void @func()
+
+declare i32 @__gcc_personality_v0(...)
+
+define void @test0() personality i8* bitcast (i32 (...)* @__gcc_personality_v0 to i8*) {
+entry:
+  invoke void @func()
+    to label %cont unwind label %lpad
+
+cont:
+  ret void
+
+lpad:
+  %exn = landingpad { i8*, i32 }
+           cleanup
+  resume { i8*, i32 } %exn
+}
+
+; IOS: __Unwind_SjLj_Resume
+; WATCHABI: __Unwind_Resume
+; EABI: _Unwind_Resume
+; ABI: _Unwind_Resume