[NAOT] GC hang on IReferenceTrackerHost_DisconnectUnusedReferenceSources with the built-in DispatcherQueueSynchronizationContext

[ahmed605]

GC hangs when XAML calls ReferenceTrackerHost.IReferenceTrackerHost_DisconnectUnusedReferenceSources under Native AOT when the built-in Windows.System.DispatcherQueueSynchronizationContext is used.

The hang only happens under NAOT and not under CoreCLR JIT (even with NAOT emulation mode), it also seems to happen only when the built-in DispatcherQueueSynchronizationContext is used, I tried with a custom implementation based on this PR and with another implementation based on the same PR but uses CoreDispatcher rather than DispatcherQueue and both had no hangs.

TTD trace: https://1drv.ms/u/c/698e7f4810a15314/IQCt7gw4FC_ISYducKW8evBoAS7MZVCJsQ87y1GNB9ba0VM?e=rzMJMY
PDB (in case it's needed): https://1drv.ms/u/c/698e7f4810a15314/IQARY-TE2ZQPTb0Bf_WVubokAREiqnst1212Y4RfK-LWjwI?e=L4emh3

Main thread:

Finalizer thread:

Environment

SDK: .NET 10.0.102 SDK
OS: Windows 11 build 22631.5472
TFM: net10.0-windows10.0.26100.0
RID: win-x64
Windows SDK package version: 10.0.26100.81
CsWinRT version: 2.3.0-prerelease.251115.2
UseUwp: true

[Sergio0694]

cc. @manodasanW

[MihuBot]

I'm a bot. Here is a possible related and/or duplicate issue (I may be wrong):

• Hang during shutdown with Native AOT on IReferenceTrackerHost::ReleaseDisconnectedReferenceSources waiting for finalizers #109538

[ahmed605]

strange, I can't find any difference between the built-in DispatcherQueueSynchronizationContext and the one in the PR I linked above that could possibly cause this, maybe DispatcherQueueSynchronizationContext isn't really the issue but swapping it affected something else indirectly that somehow prevents the hang from happening?

[manodasanW]

Based on the stack, looks similar to #121538 which was addressed by #121558. But it is unclear how the implementation of DispatcherQueueSynchronizationContext makes a difference. @ahmed605 to confirm, is this a release AOT build and you don't have any property disabling optimizations as per in the other issue?

[ahmed605]

it's a release AOT build, and I don't have any property disabling optimizations

[ahmed605]

is that PR included in the 10.0.102 SDK? if not I can try a nightly build or a local build

[manodasanW]

I believe that PR is only .NET 11 as the assumption was it only impacted debug AOT builds and AOT builds with optimizations turned off as per the conversation in the linked issue.

[ahmed605]

@manodasanW I can confirm that the PR does indeed fix the issue (or at least something did in .NET 11), I guess I will close this issue now since it's already fixed, I still find it really strange how swapping the SynchronizationContext impl fixed the hang on .NET 10 though

[Sergio0694]

@jkotas @AaronRobinsonMSFT should we consider backporting the fix to .NET 10, if the bug causes deadlock in applications even if they're compiled with optimizations enabled? My understanding is that the fix wasn't backported mainly because it was thought to only impact apps built with optimizations disabled (which isn't supported), but it seems like it's not actually the case? 🤔

[ahmed605]

if that helps this is my publish profile:

<?xml version="1.0" encoding="utf-8"?>
<!--
https://go.microsoft.com/fwlink/?LinkID=208121.
-->
<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
  <PropertyGroup>
    <PublishProtocol>FileSystem</PublishProtocol>
    <Platform>x64</Platform>
    <RuntimeIdentifier>win-x64</RuntimeIdentifier>
    <PublishDir>bin\$(Configuration)\$(TargetFramework)\$(RuntimeIdentifier)\publish\</PublishDir>
    <SelfContained>true</SelfContained>
  </PropertyGroup>
</Project>

and these are all properties set in my csproj:

<PropertyGroup>
  <OutputType>WinExe</OutputType>
  <TargetFramework>net10.0-windows10.0.26100.0</TargetFramework>
  <WindowsSdkPackageVersion>10.0.26100.81</WindowsSdkPackageVersion>
  <TargetPlatformVersion>10.0.26100.0</TargetPlatformVersion>
  <ImplicitUsings>enable</ImplicitUsings>
  <Nullable>enable</Nullable>
  <PublishAot>true</PublishAot>
  <InvariantGlobalization>true</InvariantGlobalization>
  <DisableRuntimeMarshalling>true</DisableRuntimeMarshalling>
  <UseUwp>true</UseUwp>
  <UseUwpTools>true</UseUwpTools>
  <DefineConstants>DISABLE_XAML_GENERATED_MAIN;COREDISPATCHER_FALLBACK;$(DefineConstants)</DefineConstants>
  <AllowUnsafeBlocks>true</AllowUnsafeBlocks>
  <Platforms>x64;arm64;x86</Platforms>
  <RuntimeIdentifiers>win-x86;win-x64;win-arm64</RuntimeIdentifiers>
  <PublishProfile>win-$(Platform).pubxml</PublishProfile>
  <ApplicationManifest>app.manifest</ApplicationManifest>
  <DisableEmbeddedXbf>false</DisableEmbeddedXbf>
  <EnableMsixTooling>true</EnableMsixTooling>
  <ProjectPriFileName>resources.pri</ProjectPriFileName>
  <_SupportEmbedFileResources>true</_SupportEmbedFileResources>
  <CsWinRTAotWarningLevel>3</CsWinRTAotWarningLevel>
  <LangVersion>preview</LangVersion>
  <CsWinRTIIDOptimizerOptOut>true</CsWinRTIIDOptimizerOptOut>
</PropertyGroup>

the AOT build was created using VS through Build > Publish Selection with the Release configuration selected in VS' top bar