[asan] Keep Itanium mangled names in global metadata

[MaskRay]

Summary

The runtime calls MaybeDemangleGlobalName for error reporting and
__cxxabiv1::__cxa_demangle is called if available, so demanging Itanium
mangled names in global metadata is unnecessary and wastes data size.

Add MaybeDemangleGlobalName in ODR violation detection to support demangled
names in a suppressions file. MaybeDemangleGlobalName may call
DemangleCXXABI and leak memory. Use an internal allocation to prevent lsan
leak (in case there is no fatal asan error).

The debug feature report_globals=2 prints information for all instrumented
global variables. MaybeDemangleGlobalName would be slow, so don't do that.
The output looks like Added Global[0x56448f092d60]: beg=0x56448fa66d60 size=4/32 name=_ZL13test_global_2
and I think the mangled name is fine.

Other mangled schemes e.g. Windows (see win-string-literal.ll) remain the
current behavior.

[MaskRay]

always keep original name for __odr_gen_asan_

[hctim]

couple nits, otherwise lgtm

compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp

49–62

Can you maybe update this entire fixme?

Looks like all the current callers would be tolerant to a function-local static buffer, but this feels like a gunfoot, so I guess we'll continue to leak the memory, unless you also want to go ahead and create an InternalAlloc-RAII-wrapper class and patch up the callers.

FIXME: This pointer currently leaks in the callers, but we specifically copy the demangled contents into an InternalAlloc so that LSan doesn't see a leak from the malloc() done by cxa_demangle.

55

nit: if you're here, can you change these parameters to nullptr?

[MaskRay]

address comments

Thanks for the comment suggestion!

[dblaikie]

Huh, fascinating - so we used the unmangled name as a symbol name, punctuation, spaces, and all?

(interesting to know that sanitizers end up with their own copies of mangled names too - another thing we could consider deduplicating/maybe sharing with some debug info/symbol table reductions at some point (I guess if we just simplify mangled names enough - use hashes, etc, this'd fall out naturally for the sanitizers (though all demanglers would need to have access to some side table of hash->mangled name table, which I guess sanitizers need at runtime, so couldn't be dropped during linking (previously had considered this strategy since the linker would need to produce good diagnostics - but I guess runtime symbolizers need useful symbol names too, etc)))

[hctim]

In D138095#3931484, @dblaikie wrote:

Huh, fascinating - so we used the unmangled name as a symbol name, punctuation, spaces, and all?

(interesting to know that sanitizers end up with their own copies of mangled names too - another thing we could consider deduplicating/maybe sharing with some debug info/symbol table reductions at some point (I guess if we just simplify mangled names enough - use hashes, etc, this'd fall out naturally for the sanitizers (though all demanglers would need to have access to some side table of hash->mangled name table, which I guess sanitizers need at runtime, so couldn't be dropped during linking (previously had considered this strategy since the linker would need to produce good diagnostics - but I guess runtime symbolizers need useful symbol names too, etc)))

nice triple-nested parenthetical thought!

yes - this is a room for opportunity :). i think the major thing that would need to happen is for llvm-symbolizer to be able to symbolize an address half way through a GV (although it may be possible to find the start pointer using the sanitizer metadata).

[MaskRay]

In D138095#3931582, @hctim wrote:

In D138095#3931484, @dblaikie wrote:

Huh, fascinating - so we used the unmangled name as a symbol name, punctuation, spaces, and all?

(interesting to know that sanitizers end up with their own copies of mangled names too - another thing we could consider deduplicating/maybe sharing with some debug info/symbol table reductions at some point (I guess if we just simplify mangled names enough - use hashes, etc, this'd fall out naturally for the sanitizers (though all demanglers would need to have access to some side table of hash->mangled name table, which I guess sanitizers need at runtime, so couldn't be dropped during linking (previously had considered this strategy since the linker would need to produce good diagnostics - but I guess runtime symbolizers need useful symbol names too, etc)))

nice triple-nested parenthetical thought!

yes - this is a room for opportunity :). i think the major thing that would need to happen is for llvm-symbolizer to be able to symbolize an address half way through a GV (although it may be possible to find the start pointer using the sanitizer metadata).

Yes, definitely room for improvement. The instrumentation - runtime protocol is more private than DWARF debug information (which has more consumers and therefore has more restriction on things like string table usage.)
In some operating systems that adopt .ctf, I think .ctf just reuses the ELF symbol table .symtab and .strtab, resulting in some saving. (This is some work going on in binutils, but I don't actively keep an eye on it.)

The compression+hash idea has been deployed elsewhere in llvm-project.
In instrumentaiton-based PGO, __llvm_prf_nm uses compressed content with hashes indexing into it. If asan is happy with requiring a non-SHF_ALLOC section for symbolization: we can use a generalized --compress-debug-sections= : https://sourceware.org/bugzilla/show_bug.cgi?id=27452

[gulfem]

We started seeing the following test failures.

Failed Tests (8):
  AddressSanitizer-x86_64-linux :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location.cpp

https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64/b8797107374001425041/overview

The error is message from global-demangle.cpp is as the following:

FAIL: AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp (560 of 12394)
******************** TEST 'AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';      /b/s/w/ir/x/w/staging/llvm_build/./bin/clang  --driver-mode=g++ -fsanitize=address -mno-omit-leaf-frame-pointer -fno-omit-frame-pointer -fno-optimize-sibling-calls -gline-tables-only   -Wthread-safety -Wthread-safety-reference -Wthread-safety-beta   -O0 /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp -o /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp && not  /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp 2>&1 | FileCheck /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp
--
Exit Code: 1

Command Output (stderr):
--
/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:15:12: error: CHECK: expected string not found in input
 // CHECK: '{{.*}}XXX::YYY::ZZZ{{.*}}' {{.*}} of size 4
           ^
<stdin>:8:56: note: scanning from here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                       ^
<stdin>:8:165: note: possible intended match here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                                                                                                                                    ^

Input file: <stdin>
Check file: /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp

-dump-input=help explains the following input dump.

Input was:
<<<<<<
            1: ================================================================= 
            2: ==315196==ERROR: AddressSanitizer: global-buffer-overflow on address 0x561bdea237e6 at pc 0x561bdea15d7e bp 0x7ffda4270b00 sp 0x7ffda4270af8 
            3: READ of size 1 at 0x561bdea237e6 thread T0 
            4:  #0 0x561bdea15d7d in main /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 
            5:  #1 0x7fe9bc8a1d09 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x23d09) (BuildId: 814ca45fb917d1b757796ff5f456237c263711c7) 
            6:  #2 0x561bde940339 in _start (/b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp+0x59339) (BuildId: 4e75c435413f2362) 
            7:  
            8: 0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4 
check:15'0                                                            X~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ error: no match found
check:15'1                                                                                                                                                                         ?                                   possible intended match
            9:  '_ZN3XXX3YYY3ZZZE' is ascii string 'abc' 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           10: SUMMARY: AddressSanitizer: global-buffer-overflow /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 in main 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           11: Shadow bytes around the buggy address: 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           12:  0x561bdea23500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           13:  0x561bdea23580: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            .
            .
            .
>>>>>>

[MaskRay]

In D138095#3941776, @gulfem wrote:

We started seeing the following test failures.

Failed Tests (8):
  AddressSanitizer-x86_64-linux :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location.cpp

https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64/b8797107374001425041/overview

The error is message from global-demangle.cpp is as the following:

FAIL: AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp (560 of 12394)
******************** TEST 'AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';      /b/s/w/ir/x/w/staging/llvm_build/./bin/clang  --driver-mode=g++ -fsanitize=address -mno-omit-leaf-frame-pointer -fno-omit-frame-pointer -fno-optimize-sibling-calls -gline-tables-only   -Wthread-safety -Wthread-safety-reference -Wthread-safety-beta   -O0 /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp -o /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp && not  /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp 2>&1 | FileCheck /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp
--
Exit Code: 1

Command Output (stderr):
--
/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:15:12: error: CHECK: expected string not found in input
 // CHECK: '{{.*}}XXX::YYY::ZZZ{{.*}}' {{.*}} of size 4
           ^
<stdin>:8:56: note: scanning from here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                       ^
<stdin>:8:165: note: possible intended match here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                                                                                                                                    ^

Input file: <stdin>
Check file: /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp

-dump-input=help explains the following input dump.

Input was:
<<<<<<
            1: ================================================================= 
            2: ==315196==ERROR: AddressSanitizer: global-buffer-overflow on address 0x561bdea237e6 at pc 0x561bdea15d7e bp 0x7ffda4270b00 sp 0x7ffda4270af8 
            3: READ of size 1 at 0x561bdea237e6 thread T0 
            4:  #0 0x561bdea15d7d in main /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 
            5:  #1 0x7fe9bc8a1d09 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x23d09) (BuildId: 814ca45fb917d1b757796ff5f456237c263711c7) 
            6:  #2 0x561bde940339 in _start (/b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp+0x59339) (BuildId: 4e75c435413f2362) 
            7:  
            8: 0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4 
check:15'0                                                            X~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ error: no match found
check:15'1                                                                                                                                                                         ?                                   possible intended match
            9:  '_ZN3XXX3YYY3ZZZE' is ascii string 'abc' 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           10: SUMMARY: AddressSanitizer: global-buffer-overflow /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 in main 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           11: Shadow bytes around the buggy address: 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           12:  0x561bdea23500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           13:  0x561bdea23580: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            .
            .
            .
>>>>>>

I'll appreciate some debugging since other sanitizer bots seem fine? Does your build bot not link asan tests with libc++abi or libstdc++ which provides __cxa_demangle? Then a mangled name is expected.
I vaguely remember that Fuchsia does things somewhat differently.

[MaskRay]

Perhaps you use -static-libstdc++ for both %dynamiclib and the main executable? Then there is no defined __cxa_demangle in the process.
I wonder whether a demangled name in odr-violation diagnostics is useful... Personally whether it is mangled/demangled doesn't concern me.

[hctim]

In D138095#3941942, @MaskRay wrote:

I wonder whether a demangled name in odr-violation diagnostics is useful... Personally whether it is mangled/demangled doesn't concern me.

Given the amount of "what does this use-after-free report mean, stop bothering me" we get, I'm appreciate a demangled name because it likely reduces the amount of bug reports assigned to me ;).

[MaskRay]

In D138095#3941961, @hctim wrote:

In D138095#3941942, @MaskRay wrote:

I wonder whether a demangled name in odr-violation diagnostics is useful... Personally whether it is mangled/demangled doesn't concern me.

Given the amount of "what does this use-after-free report mean, stop bothering me" we get, I'm appreciate a demangled name because it likely reduces the amount of bug reports assigned to me ;).

Thanks for the quick reply! Then I'll probably revert this patch and not consider reinstating it.

compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp uses an undefined weak __cxa_demangle which does not pull in an archive definition.
So a -static-libstdc++ mostly-statically executable won't get demangled names.

[zequanwu]

Hi, this causes two tests failure on mac.

Failed Tests (2):
   SanitizerCommon-tsan-x86_64-Darwin :: Darwin/symbolizer-function-offset-dladdr.cpp
   ThreadSanitizer-x86_64 :: Darwin/symbolizer-dladdr.cpp

https://reviews.llvm.org/rGdb7c82231c32074440d5426870051228435c6bce doesn't seem to fix the test because our cron job picks 57fd7ffe which includes it and failed on 3 tests:

Failed Tests (3):
   AddressSanitizer-x86_64-darwin :: TestCases/Darwin/interface_symbols_darwin.cpp
   SanitizerCommon-tsan-x86_64-Darwin :: Darwin/symbolizer-function-offset-dladdr.cpp
   ThreadSanitizer-x86_64 :: Darwin/symbolizer-dladdr.cpp

See https://bugs.chromium.org/p/chromium/issues/detail?id=1392533 for details. Please revert those two commits if it takes a while to fix.

[MaskRay]

In D138095#3945190, @zequanwu wrote:

Hi, this causes two tests failure on mac.

Failed Tests (2):
   SanitizerCommon-tsan-x86_64-Darwin :: Darwin/symbolizer-function-offset-dladdr.cpp
   ThreadSanitizer-x86_64 :: Darwin/symbolizer-dladdr.cpp

https://reviews.llvm.org/rGdb7c82231c32074440d5426870051228435c6bce doesn't seem to fix the test because our cron job picks 57fd7ffe which includes it and failed on 3 tests:

Failed Tests (3):
   AddressSanitizer-x86_64-darwin :: TestCases/Darwin/interface_symbols_darwin.cpp
   SanitizerCommon-tsan-x86_64-Darwin :: Darwin/symbolizer-function-offset-dladdr.cpp
   ThreadSanitizer-x86_64 :: Darwin/symbolizer-dladdr.cpp

See https://bugs.chromium.org/p/chromium/issues/detail?id=1392533 for details. Please revert those two commits if it takes a while to fix.

I don't have a macOS to check issues but don't think the combination of 00be3578e0841dd9abe408e5b4946180de0bf46b+db7c82231c32074440d5426870051228435c6bce (which just changed a __cxa_demangle call site and the ODR indicator symbol name) can suppress a tsan failure).

compiler-rt/test/asan/TestCases/Darwin/interface_symbols_darwin.cpp looks like a test which is nearly impossible to investigate without a macOS though I don't think it has connection with this patch.

[MaskRay]

The reverse of the combined commits is the following. Any chance you found the wrong culprit?

diff --git a/compiler-rt/lib/asan/asan_globals.cpp b/compiler-rt/lib/asan/asan_globals.cpp
index 2b4a9e01c8a3..b780128c9adb 100644
--- a/compiler-rt/lib/asan/asan_globals.cpp
+++ b/compiler-rt/lib/asan/asan_globals.cpp
@@ -151,4 +151,4 @@ static void CheckODRViolationViaIndicator(const Global *g) {
         !IsODRViolationSuppressed(g->name))
-      ReportODRViolation(g, FindRegistrationSite(g), l->g,
-                         FindRegistrationSite(l->g));
+      ReportODRViolation(g, FindRegistrationSite(g),
+                         l->g, FindRegistrationSite(l->g));
   }
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp
index d505d96bd653..b223f6cd01e3 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp
@@ -51,13 +51,8 @@ const char *DemangleCXXABI(const char *name) {
   // own demangler (libc++abi's implementation could be adapted so that
-  // it does not allocate). For now, we just call it anyway, and use
-  // InternalAlloc to prevent lsan error.
-  if (&__cxxabiv1::__cxa_demangle) {
-    if (char *demangled_name = __cxxabiv1::__cxa_demangle(name, 0, 0, 0)) {
-      size_t size = internal_strlen(demangled_name) + 1;
-      char *buf = (char *)InternalAlloc(size);
-      internal_memcpy(buf, demangled_name, size);
-      free(demangled_name);
-      return buf;
-    }
-  }
+  // it does not allocate). For now, we just call it anyway, and we leak
+  // the returned value.
+  if (&__cxxabiv1::__cxa_demangle)
+    if (const char *demangled_name =
+          __cxxabiv1::__cxa_demangle(name, 0, 0, 0))
+      return demangled_name;
 
diff --git a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
index a4d4cc73c482..ff05454aa920 100644
--- a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
@@ -2266,8 +2266,7 @@ bool ModuleAddressSanitizer::InstrumentGlobals(IRBuilder<> &IRB, Module &M,
 
-    // The runtime library tries demangling symbol names in the descriptor but
-    // functionality like __cxa_demangle may be unavailable (e.g.
-    // -static-libstdc++). So we demangle the symbol names here.
-    std::string NameForGlobal = G->getName().str();
+    // TODO: Symbol names in the descriptor can be demangled by the runtime
+    // library. This could save ~0.4% of VM size for a private large binary.
+    std::string NameForGlobal = llvm::demangle(G->getName().str());
     GlobalVariable *Name =
-        createPrivateGlobalForString(M, llvm::demangle(NameForGlobal),
+        createPrivateGlobalForString(M, NameForGlobal,
                                      /*AllowMerging*/ true, kAsanGenPrefix);

[gulfem]

In D138095#3941913, @MaskRay wrote:

In D138095#3941776, @gulfem wrote:

We started seeing the following test failures.

Failed Tests (8):
  AddressSanitizer-x86_64-linux :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location.cpp

https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64/b8797107374001425041/overview

The error is message from global-demangle.cpp is as the following:

FAIL: AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp (560 of 12394)
******************** TEST 'AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';      /b/s/w/ir/x/w/staging/llvm_build/./bin/clang  --driver-mode=g++ -fsanitize=address -mno-omit-leaf-frame-pointer -fno-omit-frame-pointer -fno-optimize-sibling-calls -gline-tables-only   -Wthread-safety -Wthread-safety-reference -Wthread-safety-beta   -O0 /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp -o /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp && not  /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp 2>&1 | FileCheck /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp
--
Exit Code: 1

Command Output (stderr):
--
/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:15:12: error: CHECK: expected string not found in input
 // CHECK: '{{.*}}XXX::YYY::ZZZ{{.*}}' {{.*}} of size 4
           ^
<stdin>:8:56: note: scanning from here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                       ^
<stdin>:8:165: note: possible intended match here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                                                                                                                                    ^

Input file: <stdin>
Check file: /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp

-dump-input=help explains the following input dump.

Input was:
<<<<<<
            1: ================================================================= 
            2: ==315196==ERROR: AddressSanitizer: global-buffer-overflow on address 0x561bdea237e6 at pc 0x561bdea15d7e bp 0x7ffda4270b00 sp 0x7ffda4270af8 
            3: READ of size 1 at 0x561bdea237e6 thread T0 
            4:  #0 0x561bdea15d7d in main /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 
            5:  #1 0x7fe9bc8a1d09 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x23d09) (BuildId: 814ca45fb917d1b757796ff5f456237c263711c7) 
            6:  #2 0x561bde940339 in _start (/b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp+0x59339) (BuildId: 4e75c435413f2362) 
            7:  
            8: 0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4 
check:15'0                                                            X~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ error: no match found
check:15'1                                                                                                                                                                         ?                                   possible intended match
            9:  '_ZN3XXX3YYY3ZZZE' is ascii string 'abc' 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           10: SUMMARY: AddressSanitizer: global-buffer-overflow /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 in main 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           11: Shadow bytes around the buggy address: 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           12:  0x561bdea23500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           13:  0x561bdea23580: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            .
            .
            .
>>>>>>

I'll appreciate some debugging since other sanitizer bots seem fine? Does your build bot not link asan tests with libc++abi or libstdc++ which provides __cxa_demangle? Then a mangled name is expected.
I vaguely remember that Fuchsia does things somewhat differently.

After https://reviews.llvm.org/rGdb7c82231c32074440d5426870051228435c6bce, we do not see the test asan test failures anymore.
Are you planning to reland this patch or did you abandon it? We encountered the issue while building the Fuchsia toolchain for the host (Linux), and these tests failed on Linux, not Fuchsia.

[zequanwu]

In D138095#3945215, @MaskRay wrote:

The reverse of the combined commits is the following. Any chance you found the wrong culprit?

diff --git a/compiler-rt/lib/asan/asan_globals.cpp b/compiler-rt/lib/asan/asan_globals.cpp
index 2b4a9e01c8a3..b780128c9adb 100644
--- a/compiler-rt/lib/asan/asan_globals.cpp
+++ b/compiler-rt/lib/asan/asan_globals.cpp
@@ -151,4 +151,4 @@ static void CheckODRViolationViaIndicator(const Global *g) {
         !IsODRViolationSuppressed(g->name))
-      ReportODRViolation(g, FindRegistrationSite(g), l->g,
-                         FindRegistrationSite(l->g));
+      ReportODRViolation(g, FindRegistrationSite(g),
+                         l->g, FindRegistrationSite(l->g));
   }
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp
index d505d96bd653..b223f6cd01e3 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp
@@ -51,13 +51,8 @@ const char *DemangleCXXABI(const char *name) {
   // own demangler (libc++abi's implementation could be adapted so that
-  // it does not allocate). For now, we just call it anyway, and use
-  // InternalAlloc to prevent lsan error.
-  if (&__cxxabiv1::__cxa_demangle) {
-    if (char *demangled_name = __cxxabiv1::__cxa_demangle(name, 0, 0, 0)) {
-      size_t size = internal_strlen(demangled_name) + 1;
-      char *buf = (char *)InternalAlloc(size);
-      internal_memcpy(buf, demangled_name, size);
-      free(demangled_name);
-      return buf;
-    }
-  }
+  // it does not allocate). For now, we just call it anyway, and we leak
+  // the returned value.
+  if (&__cxxabiv1::__cxa_demangle)
+    if (const char *demangled_name =
+          __cxxabiv1::__cxa_demangle(name, 0, 0, 0))
+      return demangled_name;
 
diff --git a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
index a4d4cc73c482..ff05454aa920 100644
--- a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
@@ -2266,8 +2266,7 @@ bool ModuleAddressSanitizer::InstrumentGlobals(IRBuilder<> &IRB, Module &M,
 
-    // The runtime library tries demangling symbol names in the descriptor but
-    // functionality like __cxa_demangle may be unavailable (e.g.
-    // -static-libstdc++). So we demangle the symbol names here.
-    std::string NameForGlobal = G->getName().str();
+    // TODO: Symbol names in the descriptor can be demangled by the runtime
+    // library. This could save ~0.4% of VM size for a private large binary.
+    std::string NameForGlobal = llvm::demangle(G->getName().str());
     GlobalVariable *Name =
-        createPrivateGlobalForString(M, llvm::demangle(NameForGlobal),
+        createPrivateGlobalForString(M, NameForGlobal,
                                      /*AllowMerging*/ true, kAsanGenPrefix);

I just manually double checked. At least this change causes these two tests(symbolizer-function-offset-dladdr.cpp and symbolizer-dladdr.cpp) to fail, not sure about interface_symbols_darwin.cpp.
I have a macbook and can help with debugging the tests.

[MaskRay]

In D138095#3945279, @gulfem wrote:

In D138095#3941913, @MaskRay wrote:

In D138095#3941776, @gulfem wrote:

We started seeing the following test failures.

Failed Tests (8):
  AddressSanitizer-x86_64-linux :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux :: TestCases/global-location.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/Linux/odr-violation.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-demangle.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location-nodebug.cpp
  AddressSanitizer-x86_64-linux-dynamic :: TestCases/global-location.cpp

https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64/b8797107374001425041/overview

The error is message from global-demangle.cpp is as the following:

FAIL: AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp (560 of 12394)
******************** TEST 'AddressSanitizer-x86_64-linux :: TestCases/global-demangle.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';      /b/s/w/ir/x/w/staging/llvm_build/./bin/clang  --driver-mode=g++ -fsanitize=address -mno-omit-leaf-frame-pointer -fno-omit-frame-pointer -fno-optimize-sibling-calls -gline-tables-only   -Wthread-safety -Wthread-safety-reference -Wthread-safety-beta   -O0 /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp -o /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp && not  /b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp 2>&1 | FileCheck /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp
--
Exit Code: 1

Command Output (stderr):
--
/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:15:12: error: CHECK: expected string not found in input
 // CHECK: '{{.*}}XXX::YYY::ZZZ{{.*}}' {{.*}} of size 4
           ^
<stdin>:8:56: note: scanning from here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                       ^
<stdin>:8:165: note: possible intended match here
0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4
                                                                                                                                                                    ^

Input file: <stdin>
Check file: /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp

-dump-input=help explains the following input dump.

Input was:
<<<<<<
            1: ================================================================= 
            2: ==315196==ERROR: AddressSanitizer: global-buffer-overflow on address 0x561bdea237e6 at pc 0x561bdea15d7e bp 0x7ffda4270b00 sp 0x7ffda4270af8 
            3: READ of size 1 at 0x561bdea237e6 thread T0 
            4:  #0 0x561bdea15d7d in main /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 
            5:  #1 0x7fe9bc8a1d09 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x23d09) (BuildId: 814ca45fb917d1b757796ff5f456237c263711c7) 
            6:  #2 0x561bde940339 in _start (/b/s/w/ir/x/w/staging/llvm_build/runtimes/runtimes-x86_64-unknown-linux-gnu-bins/compiler-rt/test/asan/X86_64LinuxConfig/TestCases/Output/global-demangle.cpp.tmp+0x59339) (BuildId: 4e75c435413f2362) 
            7:  
            8: 0x561bdea237e6 is located 2 bytes after global variable '_ZN3XXX3YYY3ZZZE' defined in '/b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp' (0x561bdea237e0) of size 4 
check:15'0                                                            X~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ error: no match found
check:15'1                                                                                                                                                                         ?                                   possible intended match
            9:  '_ZN3XXX3YYY3ZZZE' is ascii string 'abc' 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           10: SUMMARY: AddressSanitizer: global-buffer-overflow /b/s/w/ir/x/w/llvm-llvm-project/compiler-rt/test/asan/TestCases/global-demangle.cpp:12:15 in main 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           11: Shadow bytes around the buggy address: 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           12:  0x561bdea23500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           13:  0x561bdea23580: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 
check:15'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
            .
            .
            .
>>>>>>

I'll appreciate some debugging since other sanitizer bots seem fine? Does your build bot not link asan tests with libc++abi or libstdc++ which provides __cxa_demangle? Then a mangled name is expected.
I vaguely remember that Fuchsia does things somewhat differently.

After https://reviews.llvm.org/rGdb7c82231c32074440d5426870051228435c6bce, we do not see the test asan test failures anymore.
Are you planning to reland this patch or did you abandon it? We encountered the issue while building the Fuchsia toolchain for the host (Linux), and these tests failed on Linux, not Fuchsia.

Thanks for checking. db7c82231c32074440d5426870051228435c6bce restored the original behavior and I do not plan to reland the demangling part of the patch (infeasible since -fsanitize=address user may not link in libc++abi/libstdc++).

[MaskRay]

Reverted it for now. Debugged a bit using @zequanwu's macOS, running a test under lldb said that free(demangled_name) in compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp segfaulted.
This is really puzzling. It's a standard practice to call free on __cxxabiv1::__cxa_demangle allocated string (realloc under the hood). Both realloc/free are intercepted by sanitizers.

[MaskRay]

Will appreciate some help why macOS doesn't like free(demangled_name) in compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp ... This doesn't feel like a code issue to me.

compiler-rt/lib/asan/asan_globals.cpp

if (*odr_indicator == UNREGISTERED) {
     *odr_indicator = REGISTERED;
     return;
   }
   // If *odr_indicator is DEFINED, some module have already registered
   // externally visible symbol with the same name. This is an ODR violation.
   for (ListOfGlobals *l = list_of_all_globals; l; l = l->next) {
     if (g->odr_indicator == l->g->odr_indicator &&
         (flags()->detect_odr_violation >= 2 || g->size != l->g->size) &&
-        !IsODRViolationSuppressed(g->name))
-      ReportODRViolation(g, FindRegistrationSite(g),
-                         l->g, FindRegistrationSite(l->g));
+        !IsODRViolationSuppressed(MaybeDemangleGlobalName(g->name)))
+      ReportODRViolation(g, FindRegistrationSite(g), l->g,
+                         FindRegistrationSite(l->g));
   }
 }
 
 // Check ODR violation for given global G by checking if it's already poisoned.
 // We use this method in case compiler doesn't use private aliases for global
 // variables.
 static void CheckODRViolationViaPoisoning(const Global *g) {
   if (__asan_region_is_poisoned(g->beg, g->size_with_redzone)) {
     // This check may not be enough: if the first global is much larger
     // the entire redzone of the second global may be within the first global.
     for (ListOfGlobals *l = list_of_all_globals; l; l = l->next) {
       if (g->beg == l->g->beg &&
           (flags()->detect_odr_violation >= 2 || g->size != l->g->size) &&
-          !IsODRViolationSuppressed(g->name))
+          !IsODRViolationSuppressed(MaybeDemangleGlobalName(g->name)))
         ReportODRViolation(g, FindRegistrationSite(g),
                            l->g, FindRegistrationSite(l->g));
     }
   }
 }
 
 // Clang provides two different ways for global variables protection:
 // it can poison the global itself or its private alias. In former

compiler-rt/lib/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cpp

namespace __cxxabiv1 {
   char *__cxa_demangle(const char *mangled, char *buffer,
                                   size_t *length, int *status);
 }
 
 namespace __sanitizer {
 
 // Attempts to demangle the name via __cxa_demangle from __cxxabiv1.
 const char *DemangleCXXABI(const char *name) {
   // FIXME: __cxa_demangle aggressively insists on allocating memory.
   // There's not much we can do about that, short of providing our
   // own demangler (libc++abi's implementation could be adapted so that
-  // it does not allocate). For now, we just call it anyway, and we leak
-  // the returned value.
-  if (&__cxxabiv1::__cxa_demangle)
-    if (const char *demangled_name =
-          __cxxabiv1::__cxa_demangle(name, 0, 0, 0))
-      return demangled_name;
+  // it does not allocate). For now, we just call it anyway, and use
+  // InternalAlloc to prevent lsan error.
+  if (&__cxxabiv1::__cxa_demangle) {
+    if (char *demangled_name = __cxxabiv1::__cxa_demangle(name, 0, 0, 0)) {
+      size_t size = internal_strlen(demangled_name) + 1;
+      char *buf = (char *)InternalAlloc(size);
+      internal_memcpy(buf, demangled_name, size);
+      free(demangled_name);
+      return buf;
+    }
+  }
 
   return name;
 }
 
 // As of now, there are no headers for the Swift runtime. Once they are
 // present, we will weakly link since we do not require Swift runtime to be
 // linked.
 typedef char *(*swift_demangle_ft)(const char *mangledName,

compiler-rt/test/asan/TestCases/Linux/odr_indicators.cpp

 // RUN: %clangxx_asan -fno-sanitize-address-use-odr-indicator -fPIC %s -o %t
 // RUN: %env_asan_opts=report_globals=2 %run %t 2>&1 | FileCheck %s --check-prefixes=CHECK,INDICATOR0
 
 // RUN: %clangxx_asan -fsanitize-address-use-odr-indicator -fPIC %s -o %t
 // RUN: %env_asan_opts=report_globals=2 %run %t 2>&1 | FileCheck %s --check-prefixes=CHECK,INDICATOR1
 
 #include <stdio.h>
 
 int test_global_1;
 // INDICATOR0-DAG: Added Global{{.*}} name=test_global_1{{.*}} odr_indicator={{0x0+$}}
 // INDICATOR1-DAG: Added Global{{.*}} name=test_global_1{{.*}} odr_indicator={{0x0*[^0]+.*$}}
 
 static int test_global_2;
-// CHECK-DAG: Added Global{{.*}} name=test_global_2{{.*}} odr_indicator={{0xf+$}}
+// CHECK-DAG: Added Global{{.*}} name=_ZL13test_global_2 {{.*}} odr_indicator={{0xf+$}}
 
 namespace {
 static int test_global_3;
-// CHECK-DAG: Added Global{{.*}} name={{.*}}::test_global_3{{.*}} odr_indicator={{0xf+$}}
+// CHECK-DAG: Added Global{{.*}} name=_ZN12_GLOBAL__N_113test_global_3E {{.*}} odr_indicator={{0xf+$}}
 } // namespace
 
 int main() {
   const char f[] = "%d %d %d\n";
   // CHECK-DAG: Added Global{{.*}} name=__const.main.f{{.*}} odr_indicator={{0xf+$}}
   printf(f, test_global_1, test_global_2, test_global_3);
   return 0;
 }

llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp

bool ModuleAddressSanitizer::InstrumentGlobals(IRBuilder<> &IRB, Module &M,
 
   for (size_t i = 0; i < n; i++) {
     GlobalVariable *G = GlobalsToChange[i];
 
     GlobalValue::SanitizerMetadata MD;
     if (G->hasSanitizerMetadata())
       MD = G->getSanitizerMetadata();
 
-    // TODO: Symbol names in the descriptor can be demangled by the runtime
-    // library. This could save ~0.4% of VM size for a private large binary.
-    std::string NameForGlobal = llvm::demangle(G->getName().str());
+    // ASan runtime demangles Itanium mangled names, so keep the original name
+    // to prevent unneeded size increase of the string table.
+    std::string NameForGlobal = G->getName().str();
+    if (!StringRef(NameForGlobal).startswith("_Z"))
+      NameForGlobal = llvm::demangle(NameForGlobal);
     GlobalVariable *Name =
         createPrivateGlobalForString(M, NameForGlobal,
                                      /*AllowMerging*/ true, kAsanGenPrefix);
 
     Type *Ty = G->getValueType();
     const uint64_t SizeInBytes = DL.getTypeAllocSize(Ty);
     const uint64_t RightRedzoneSize = getRedzoneSizeForGlobal(SizeInBytes);
     Type *RightRedZoneTy = ArrayType::get(IRB.getInt8Ty(), RightRedzoneSize);

llvm/test/Instrumentation/AddressSanitizer/local_alias.ll

 ; Defaults
 ; RUN: opt < %s -passes=asan -S | FileCheck %s --check-prefixes=CHECK-ALIAS,CHECK-INDICATOR
 
 ; {newPM,legacyPM} x {alias0,alias1} x {odr0,odr1}
 ; RUN: opt < %s -passes=asan             -asan-use-private-alias=0 -asan-use-odr-indicator=0 -S | FileCheck %s --check-prefixes=CHECK-NOALIAS,CHECK-NOINDICATOR
 ; RUN: opt < %s -passes=asan             -asan-use-private-alias=1 -asan-use-odr-indicator=0 -S | FileCheck %s --check-prefixes=CHECK-ALIAS,CHECK-NOINDICATOR
 ; RUN: opt < %s -passes=asan             -asan-use-private-alias=0 -asan-use-odr-indicator=1 -S | FileCheck %s --check-prefixes=CHECK-NOALIAS,CHECK-INDICATOR
 ; RUN: opt < %s -passes=asan             -asan-use-private-alias=1 -asan-use-odr-indicator=1 -S | FileCheck %s --check-prefixes=CHECK-ALIAS,CHECK-INDICATOR
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 @a = dso_local global [2 x i32] zeroinitializer, align 4
 @b = private global [2 x i32] zeroinitializer, align 4
 @c = internal global [2 x i32] zeroinitializer, align 4
-@d = unnamed_addr global [2 x i32] zeroinitializer, align 4
+@_ZL1d = unnamed_addr global [2 x i32] zeroinitializer, align 4
 
 ; Check that we generate internal alias and odr indicator symbols for global to be protected.
 ; CHECK-NOINDICATOR-NOT: __odr_asan_gen_a
 ; CHECK-NOALIAS-NOT: private alias
+; CHECK-INDICATOR: @___asan_gen_.1 = private unnamed_addr constant [2 x i8] c"a\00", align 1
 ; CHECK-INDICATOR: @__odr_asan_gen_a = global i8 0, align 1
+; CHECK-INDICATOR: @___asan_gen_.4 = private unnamed_addr constant [6 x i8] c"_ZL1d\00", align 1
+; CHECK-INDICATOR: @__odr_asan_gen__ZL1d = global i8 0, align 1
 ; CHECK-ALIAS: @0 = private alias { [2 x i32], [24 x i8] }, ptr @a
 
 ; CHECK-ALIAS: @1 = private alias { [2 x i32], [24 x i8] }, ptr @b
 ; CHECK-ALIAS: @2 = private alias { [2 x i32], [24 x i8] }, ptr @c
-; CHECK-ALIAS: @3 = private alias { [2 x i32], [24 x i8] }, ptr @d
+; CHECK-ALIAS: @3 = private alias { [2 x i32], [24 x i8] }, ptr @_ZL1d
 
 ; Function Attrs: nounwind sanitize_address uwtable
 define i32 @foo(i32 %M) #0 {
 entry:
   %M.addr = alloca i32, align 4
   store i32 %M, ptr %M.addr, align 4
   store volatile i32 6, ptr getelementptr inbounds ([2 x i32], ptr @a, i64 2, i64 0), align 4
   %0 = load i32, ptr %M.addr, align 4
   %idxprom = sext i32 %0 to i64
   %arrayidx = getelementptr inbounds [2 x i32], ptr @a, i64 0, i64 %idxprom
   %1 = load volatile i32, ptr %arrayidx, align 4
   ret i32 %1
 }