[lsan] Move allocator base to avoid conflict with high-entropy ASLR for x86-64 Linux

[MaskRay]

Summary

This ports D148280 for ASan.

On x86-64 Linux, when mmap_rnd_bits is set to 32 (the maximum ARCH_MMAP_RND_BITS_MAX),
the allocator space [kAllocatorSpace,kAllocatorSpace+kAllocatorSize)
collides with the PIE load base range (0x555555554000 upto 2**mmap_rnd_bits * pagesize away),
which can cause the allocation to fail.
Using 0x500000000000ULL as the base address avoids this problem and works with
AArch64 Linux and FreeBSD as well.

While here, change s390x to use 0x500000000000ULL as well. See D78644
discussions that the address works.

[thurston]

D147984 got reverted because it broke Apple (oops, my bad). It might be worthwhile either gating this change to non-Apple (as suggested by lgrey for ASan), or holding off on this patch until it's clear what happened with D147984.

[MaskRay]

Special case SANITIZER_APPLE

I asked a question regarding s390x: D78644

[MaskRay]

remove special case for s390x

[paulkirth]

Hi, we're seeing some ASAN test failures after this patch. Maybe the issue is the change of the condition in the preprocessor? Could you take a look, and revert if its not easy to fix?
Failing Tests:
AddressSanitizer-aarch64-linux :: TestCases/use-after-scope-capture.cpp
AddressSanitizer-aarch64-linux-dynamic :: TestCases/use-after-scope-capture.cpp
AddressSanitizer-x86_64-linux :: TestCases/use-after-scope-capture.cpp
AddressSanitizer-x86_64-linux-dynamic :: TestCases/use-after-scope-capture.cpp

https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64/b8783819018697822145/overview
https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-arm64/b8783820264168551889/overview
https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64-rbe/b8783818710722598305/overview

[MaskRay]

In D148193#4269471, @paulkirth wrote:

Hi, we're seeing some ASAN test failures after this patch. Maybe the issue is the change of the condition in the preprocessor? Could you take a look, and revert if its not easy to fix?
Failing Tests:
AddressSanitizer-aarch64-linux :: TestCases/use-after-scope-capture.cpp
AddressSanitizer-aarch64-linux-dynamic :: TestCases/use-after-scope-capture.cpp
AddressSanitizer-x86_64-linux :: TestCases/use-after-scope-capture.cpp
AddressSanitizer-x86_64-linux-dynamic :: TestCases/use-after-scope-capture.cpp

https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64/b8783819018697822145/overview
https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-arm64/b8783820264168551889/overview
https://luci-milo.appspot.com/ui/p/fuchsia/builders/toolchain.ci/clang-linux-x64-rbe/b8783818710722598305/overview

I am afraid you likely found the wrong culprit. This patch is about lsan.

TEST 'AddressSanitizer-x86_64-linux-dynamic :: TestCases/use-after-scope-capture.cpp' FAILED indicates an issue in asan (not the lsan part).

I synced to one commit before this one and the tests still fail.

[thurston]

Based on the redness for the Windows builds at at http://google.github.io/sanitizers/show_bots.html starting from https://lab.llvm.org/buildbot/#/builders/127/builds/46723
I'm guessing the commit that caused is https://github.com/llvm/llvm-project/commit/5b386b864c7619897c51a1da97d78f1cf6f3eff6

I'm currently rebuilding to test it

[thurston]

I rebuilt 0b911a3dc34a2a9930c8156183af3270f8907f5a (the commit immediately prior to the one that caused the failure) and confirmed that the test passes.

I haven't dug deep enough to find out whether it is a bug in the compiler, AddressSanitizer, or the test.

[thurston]

I've pinged the sanitizer build watcher

[paulkirth]

@MaskRay sorry for the noise. I was under the impression that LSAN was enabled by default for ASAN on Linux unless you passed additional args.

@thurston thanks for digging in.

[thurston]

The test has been fixed by fmayer in https://reviews.llvm.org/D148376

compiler-rt/lib/lsan/lsan_allocator.h

 };
 template <typename AddressSpaceView>
 using PrimaryAllocatorASVT = SizeClassAllocator32<AP32<AddressSpaceView>>;
 using PrimaryAllocator = PrimaryAllocatorASVT<LocalAddressSpaceView>;
 #else
 # if SANITIZER_FUCHSIA || defined(__powerpc64__)
 const uptr kAllocatorSpace = ~(uptr)0;
 const uptr kAllocatorSize  =  0x40000000000ULL;  // 4T.
-#elif defined(__s390x__)
-const uptr kAllocatorSpace = 0x40000000000ULL;
-const uptr kAllocatorSize = 0x40000000000ULL;  // 4T.
-# else
-const uptr kAllocatorSpace = 0x600000000000ULL;
-const uptr kAllocatorSize  = 0x40000000000ULL;  // 4T.
-# endif
+#  elif SANITIZER_APPLE
+const uptr kAllocatorSpace = 0x600000000000ULL;
+const uptr kAllocatorSize  = 0x40000000000ULL;  // 4T.
+#  else
+const uptr kAllocatorSpace = 0x500000000000ULL;
+const uptr kAllocatorSize = 0x40000000000ULL;  // 4T.
+#  endif
 template <typename AddressSpaceViewTy>
 struct AP64 {  // Allocator64 parameters. Deliberately using a short name.
   static const uptr kSpaceBeg = kAllocatorSpace;
   static const uptr kSpaceSize = kAllocatorSize;
   static const uptr kMetadataSize = sizeof(ChunkMetadata);
   typedef DefaultSizeClassMap SizeClassMap;
   typedef NoOpMapUnmapCallback MapUnmapCallback;
   static const uptr kFlags = 0;