Tasks issues on Mac Catalyst

[directhex]

Description

* thread #11, name = 'com.apple.NSEventThread', stop reason = EXC_BAD_ACCESS (code=1, address=0x2ee00)
    frame #0: 0x00007fff20243ed1 libsystem_malloc.dylib`tiny_malloc_from_free_list + 162
libsystem_malloc.dylib`tiny_malloc_from_free_list:
->  0x7fff20243ed1 <+162>: movq   %rax, (%r9)
    0x7fff20243ed4 <+165>: jmp    0x7fff20243f9d            ; <+366>
    0x7fff20243ed9 <+170>: movq   0x828(%r15), %rax
    0x7fff20243ee0 <+177>: shrq   %cl, %rax
Target 0: (System.Threading.Tasks.Tests) stopped.

Approx the same failure with System.Threading.Tasks.Dataflow.Tests

Configuration

./build.sh -ci -arch x64 -os MacCatalyst -s libs.tests -c Release /p:ArchiveTests=true /p:MonoForceInterpreter=true /p:CrossBuild=false

Regression?

No

Other information

Related PR: #51139

[steveisok]

/cc @lambdageek

[lambdageek]

Fails in various ways, but the upshot seems to be that a LowLevelLifoSemaphore (probably s_semaphore in PortableThreadPool.WorkerThread is not initialized when methods are called on it.

Sometimes we get a null sem_ptr in

gint32
ves_icall_System_Threading_LowLevelLifoSemaphore_TimedWaitInternal (gpointer sem_ptr, gint32 timeout_ms)

Sometimes instead the TimedWait throws a NRE, or sometimes there's an access violation.

It seems like PortableThreadPool.WorkerThread.WorkerThreadStart sees a null s_semaphore sometimes. Not clear how that can happen - or why it only happens on catalyst.

[lambdageek]

This is failing because there are two tests disabled with [ActiveIssue("...", typeof(PlatformDetection), nameof(PlatformDetection.IsMonoInterpreter))]

runtime/src/libraries/System.Threading.Tasks.Dataflow/tests/Dataflow/DataflowBlockExtensionTests.cs

Line 1125 in 2e08554

[ActiveIssue("https://github.com/dotnet/runtime/issues/38817", typeof(PlatformDetection), nameof(PlatformDetection.IsMonoInterpreter))]

and

runtime/src/libraries/System.Threading.Tasks.Dataflow/tests/Dataflow/DataflowBlockExtensionTests.cs

Line 1927 in 2e08554

[ActiveIssue("https://github.com/dotnet/runtime/issues/38817", typeof(PlatformDetection), nameof(PlatformDetection.IsMonoInterpreter))]

The problem is that GetIsRunningOnMonoInterpreter is wrong on mobile (including catalyst) - we don't set the MONO_ENV_OPTIONS variable to turn on the interpreter, we use the embedding API.

runtime/src/libraries/Common/tests/TestUtilities/System/PlatformDetection.cs

Lines 467 to 477 in 2e08554

	private static bool GetIsRunningOnMonoInterpreter()
	{
	#if NETCOREAPP
	if (IsBrowser)
	return RuntimeFeature.IsDynamicCodeSupported;
	#endif
	// This is a temporary solution because mono does not support interpreter detection
	// within the runtime.
	var val = Environment.GetEnvironmentVariable("MONO_ENV_OPTIONS");
	return (val != null && val.Contains("--interpreter"));
	}

So the fix is to do correct interpreter detection.

[danmoseley]

@lambdageek I happened to notice that this issue is closed but we are still excluding tests against it

runtime/src/libraries/tests.proj

Line 263 in 8a498cc

<!-- Crashes randomly during test runs https://github.com/dotnet/runtime/issues/52460 -->