[asan] Remove confusing workaround

[vitalybuka]

Summary

The goal is to identify the bot and try to fix it.

SetSoftRssLimitExceededCallback is AsanInitInternal as I assume
that only MaybeStartBackgroudThread needs to be delayed to constructors.
Later I want to move MaybeStartBackgroudThread call into sanitizer_common.

If it needs to be reverted please provide to more info, like bot, or details about setup.

[DavidSpickett]

This caused 3 tests to fail on our Thumbv7 Linux bot https://lab.llvm.org/buildbot/#/builders/170/builds/1037, bisect confirms it.

FAIL: AddressSanitizer-armhf-linux::quarantine_size_mb.cpp
FAIL: AddressSanitizer-armhf-linux-dynamic::quarantine_size_mb.cpp
FAIL: SanitizerCommon-asan-armhf-Linux:: hard_rss_limit_mb_test.cpp

We have several bots testing 32 bit Arm, I don't see the failures on the pure Arm ones. The failure reproduces on armv7 only hardware or on an AArch64 machine running a 32 bit container.

I haven't looked into why it was added in the first place but then again I don't know a lot about sanitizer internals. Does seem odd that Arm only code running in the same container is fine.

It fails on the first RUN line after compiling the exe, I can get the full failure output if that would help.

[vitalybuka]

So it's https://lab.llvm.org/buildbot/#/builders/clang-thumbv7-full-2stage
Thanks for the info. I will restore and try to guess what can be done.

In D114934#3176654, @DavidSpickett wrote:

This caused 3 tests to fail on our Thumbv7 Linux bot https://lab.llvm.org/buildbot/#/builders/170/builds/1037, bisect confirms it.

FAIL: AddressSanitizer-armhf-linux::quarantine_size_mb.cpp
FAIL: AddressSanitizer-armhf-linux-dynamic::quarantine_size_mb.cpp
FAIL: SanitizerCommon-asan-armhf-Linux:: hard_rss_limit_mb_test.cpp

We have several bots testing 32 bit Arm, I don't see the failures on the pure Arm ones. The failure reproduces on armv7 only hardware or on an AArch64 machine running a 32 bit container.

I haven't looked into why it was added in the first place but then again I don't know a lot about sanitizer internals. Does seem odd that Arm only code running in the same container is fine.

It fails on the first RUN line after compiling the exe, I can get the full failure output if that would help.

[DavidSpickett]

FYI I've moved the bot to the silent because it's been failing for a while now. (not your mistake, we took a while to respond) You can find the new builds here: https://lab.llvm.org/staging/#/builders/162

We're squeezed for hardware and a 2 stage build is a lengthy process but you should get results eventually. If you need any more info feel free to ping me.

[vitalybuka]

In D114934#3183035, @DavidSpickett wrote:

FYI I've moved the bot to the silent because it's been failing for a while now. (not your mistake, we took a while to respond) You can find the new builds here: https://lab.llvm.org/staging/#/builders/162

We're squeezed for hardware and a 2 stage build is a lengthy process but you should get results eventually. If you need any more info feel free to ping me.

Thanks, I will check that bot. I hope after https://reviews.llvm.org/rGd71775cd2ab1 it will get back. Sorry for the distraction.
BTW. The first stage build should benefit from -DLLVM_CCACHE_BUILD=ON

[DavidSpickett]

BTW. The first stage build should benefit from -DLLVM_CCACHE_BUILD=ON

Believe it or not, we do have ccache-ing but we do it by setting the stage 1 compiler to be a script that calls ccache. (not exactly sure why we did it that way) Thanks for the heads up anyway!

compiler-rt/lib/asan/asan_rtl.cpp

void PrintAddressSpaceLayout() {
   Printf("SHADOW_OFFSET: 0x%zx\n", (uptr)SHADOW_OFFSET);
   CHECK(SHADOW_SCALE >= 3 && SHADOW_SCALE <= 7);
   if (kMidMemBeg)
     CHECK(kMidShadowBeg > kLowShadowEnd &&
           kMidMemBeg > kMidShadowEnd &&
           kHighShadowBeg > kMidMemEnd);
 }
 
-#if defined(__thumb__) && defined(__linux__)
-#define START_BACKGROUND_THREAD_IN_ASAN_INTERNAL
-#endif
-
-#ifndef START_BACKGROUND_THREAD_IN_ASAN_INTERNAL
 static bool UNUSED __local_asan_dyninit = [] {
   MaybeStartBackgroudThread();
-  SetSoftRssLimitExceededCallback(AsanSoftRssLimitExceededCallback);
-
   return false;
 }();
-#endif
 
 static void AsanInitInternal() {
   if (LIKELY(asan_inited)) return;
   SanitizerToolName = "AddressSanitizer";
   CHECK(!asan_init_is_running && "ASan init calls itself!");
   asan_init_is_running = true;
 
   CacheBinaryName();
static void AsanInitInternal() {
 
   AsanTSDInit(PlatformTSDDtor);
   InstallDeadlySignalHandlers(AsanOnDeadlySignal);
 
   AllocatorOptions allocator_options;
   allocator_options.SetFrom(flags(), common_flags());
   InitializeAllocator(allocator_options);
 
-#ifdef START_BACKGROUND_THREAD_IN_ASAN_INTERNAL
-  MaybeStartBackgroudThread();
   SetSoftRssLimitExceededCallback(AsanSoftRssLimitExceededCallback);
-#endif
 
   // On Linux AsanThread::ThreadStart() calls malloc() that's why asan_inited
   // should be set to 1 prior to initializing the threads.
   asan_inited = 1;
   asan_init_is_running = false;
 
   if (flags()->atexit)
     Atexit(asan_atexit);