Replace debugger PAL waits with cross-platform wait types

[jkoritzinsky]

The debugger's remaining multi-object PAL waits couple debugger-runtime communication to legacy PAL wait handles on non-Windows platforms. This replaces those uses with debugger-owned, cross-platform wait types and keeps native handles confined to Windows.

Design

• Add WaitHandle, WaitEvent, WaitLatch, and Windows-only NativeHandle to the debug PAL.
• Implement Unix events and latches with nonblocking anonymous pipes and poll; latches remain signaled after their first Set.
• Migrate debugger event and helper-thread waits to the typed API.
• Tie transport-session opening directly to its state predicate with a shared minipal condition variable instead of using a manual reset event (condition var is a better primitive for the abstraction and Windows has one since Vista).
• Add shared minipal recursive/non-recursive mutex and condition-variable support, and use it for System.Native LowLevelMonitor.

The Unix process-exit path intentionally retains the established polling model. This keeps process lifetime ownership in DbgTransportTarget while making the published exit notification sticky for every debugger waiter.

I looked at adding support for pidfd_open or kqueue based process waits, but the volume of code to support these platform-specific pathways for a single caller was not worth the cost compared to the wait thread + latch model.

Validation

• Windows x64 Debug CoreCLR build
• Linux x64 Debug CoreCLR build on directly hosted Ubuntu WSL
• Private diagnostic Attach.AttachedTest and Attach.BreakTest on Windows and Ubuntu
• Unix WaitEvent/WaitLatch semantics harness, 20/20 runs
• System.Native LowLevelMonitor harness, 100 iterations

Fixes#132149

Note

This PR description was generated by GitHub Copilot.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

This PR modernizes debugger/runtime waiting on non-Windows platforms by replacing remaining multi-object PAL wait usage with debugger-owned, cross-platform wait primitives, and by adding shared minipal mutex/condition-variable support that is then used by System.Native’s LowLevelMonitor.

Changes:

• Introduces debugger PAL WaitHandle/WaitEvent/WaitLatch (and Windows-only NativeHandle) with platform-specific wait implementations (Win32 waits; Unix pipes + poll).
• Adds minipal condition variable support and a distinct non-recursive mutex type, and migrates System.Native LowLevelMonitor to use them.
• Migrates multiple debugger and transport wait paths away from legacy PAL multi-waits, including transport session open state synchronization via a condition variable.

Reviewed changes

Copilot reviewed 30 out of 30 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/native/minipal/mutex.h	Adds non-recursive mutex type declarations.
src/native/minipal/mutex.c	Implements non-recursive mutex init/enter/leave/destroy.
src/native/minipal/minipalconfig.h.in	Adds HAVE_PTHREAD_CONDATTR_SETCLOCK feature define.
src/native/minipal/configure.cmake	Detects pthread library location and pthread_condattr_setclock.
src/native/minipal/conditionvariable.h	Adds minipal condition variable API surface.
src/native/minipal/conditionvariable.c	Implements minipal condition variables (Win32 + pthread).
src/native/minipal/CMakeLists.txt	Adds condition variable source to minipal build.
src/native/libs/System.Native/pal_threading.c	Switches LowLevelMonitor to minipal mutex/condvar.
src/libraries/System.Private.CoreLib/src/System/Threading/WaitSubsystem.Unix.cs	Updates comments to reflect new native minipal monitor backing.
src/coreclr/inc/cordebug.idl	Clarifies ICorDebugProcess::GetHandle behavior on Unix vs Windows.
src/coreclr/debug/shared/dbgtransportsession.cpp	Replaces event-based session-open signaling with condition variable; migrates wait usage.
src/coreclr/debug/inc/debugwait.h	Introduces debugger-owned wait handle abstractions.
src/coreclr/debug/inc/dbgtransportsession.h	Updates transport interfaces and state to use new wait types and condition variable.
src/coreclr/debug/ee/rcthread.cpp	Migrates helper-thread waits to WaitHandle::Wait and wait-type API.
src/coreclr/debug/ee/debugger.h	Updates favor events and RC thread fields to WaitEvent*.
src/coreclr/debug/ee/debugger.cpp	Migrates favor event initialization to new wait types.
src/coreclr/debug/di/windowspipeline.cpp	Returns process wait handle via debugger wait abstraction.
src/coreclr/debug/di/shimremotedatatarget.cpp	Updates transport acquisition to return a WaitHandle* and manages ownership.
src/coreclr/debug/di/rspriv.h	Switches stored process “handle” to WaitHandle* and introduces wait-handle accessor.
src/coreclr/debug/di/remoteeventchannel.cpp	Adjusts event-ack handle types to WaitEvent* and updates transport acquisition ownership.
src/coreclr/debug/di/process.cpp	Migrates multiple wait sites to WaitHandle::Wait and replaces LSEA with WaitEvent*.
src/coreclr/debug/di/nativepipeline.h	Updates native pipeline process handle contract to WaitHandle*.
src/coreclr/debug/di/localeventchannel.cpp	Wraps event-ack handle in WaitEvent for waiting via the new API.
src/coreclr/debug/di/eventchannel.h	Updates event channel API to return WaitEvent* ack handles.
src/coreclr/debug/di/dbgtransportpipeline.cpp	Migrates transport pipeline multi-waits to WaitHandle::Wait; tracks synthetic exit event delivery.
src/coreclr/debug/di/dbgtransportmanager.h	Updates transport manager to return WaitHandle** for process-exit waiting.
src/coreclr/debug/di/dbgtransportmanager.cpp	Implements Unix latch-based process exit signaling and Windows handle wrapping.
src/coreclr/debug/debug-pal/win/wait.cpp	Implements wait primitives atop Win32 handles and WaitForMultipleObjectsEx.
src/coreclr/debug/debug-pal/unix/wait.cpp	Implements wait primitives with nonblocking pipes + poll and refcounted waitables.
src/coreclr/debug/debug-pal/CMakeLists.txt	Builds/links new wait sources and configures Unix HAVE_PIPE2 compile defs.

Suppressed comments (1)

src/coreclr/debug/di/process.cpp:3019

• WaitEvent detachSetThreadContextNeededEvent(m_detachSetThreadContextNeededEvent); has the same Unix hazard as other WaitEvent(existingHandle) sites: if m_detachSetThreadContextNeededEvent is an opaque pointer HANDLE on Unix, this will select WaitEvent(bool) and create a fresh event instead of duplicating the existing one.

After adding a Unix importing ctor/factory for WaitEvent, update this site to use it so the wait set observes the real detach notification.

#ifdef OUT_OF_PROCESS_SETTHREADCONTEXT
WaitEvent detachSetThreadContextNeededEvent(m_detachSetThreadContextNeededEvent);
const WaitHandle *waitSet[] = {
&detachSetThreadContextNeededEvent, // Signaled on every debug event after the first SendCanDetach request
UnsafeGetProcessWaitHandle() // Signaled when the process exits

[Copilot]

Pull request overview

Copilot reviewed 30 out of 30 changed files in this pull request and generated 1 comment.

Suppressed comments (2)

src/coreclr/debug/inc/dbgtransportsession.h:654

• m_fInitSessionStateCondition is not default-initialized (DbgTransportSession uses a defaulted ctor), but it is read in the destructor and only set on the successful init path. If Init() returns early (e.g., condition-variable init failure), the destructor can call minipal_condition_variable_destroy on uninitialized storage.

 minipal_condition_variable m_sessionStateCondition;
bool m_fInitSessionStateCondition;

src/coreclr/debug/inc/dbgtransportsession.h:693

• m_rghEventReadyEvent is deleted in the destructor, but the array elements are not default-initialized (DbgTransportSession has a defaulted ctor). If Init() fails before setting these pointers, the destructor can end up deleting indeterminate pointer values.

 WaitEvent *m_rghEventReadyEvent[IPCET_Max]; // The event signalled when a new event arrives

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

Note

This error may be related to your runner configuration. You can now configure runners for Copilot code review separately from Copilot cloud agent by creating a copilot-code-review.yml file with your setup steps. Read the docs for details.

[noahfalk]

This looks reasonable to me. If there were subtle changes in the synchronization its unlikely I'd catch it but if debugger tests are passing I'm fine to take that risk and deal with it if anything pops up.

[Copilot]

Pull request overview

Copilot reviewed 30 out of 30 changed files in this pull request and generated no new comments.

Suppressed comments (2)

src/coreclr/debug/di/dbgtransportmanager.cpp:137

• The initial Unix liveness probe kill(pid, 0) does not handle EINTR. A signal delivery can cause a spurious failure to attach, even though the process exists. The poller thread already handles EINTR; the probe should match that behavior.

#ifdef HOST_UNIX
if (kill(dwPID, 0) != 0)
{
transport->Shutdown();
return (errno == ESRCH) ? E_INVALIDARG : E_FAIL;

src/coreclr/debug/shared/dbgtransportsession.cpp:748

• In the failure path for duplicating the reply waitable, this returns a generic E_FAIL. On Windows this can lose the underlying Win32 error from DuplicateHandle, making it hard to diagnose (and inconsistent with the earlier behavior that surfaced the OS error). Capture GetLastError() before deleting the original event and return HRESULT_FROM_WIN32(error) on Windows.

 WaitEvent hReplyEvent(*pMessage->m_hReplyEvent);
if (!hReplyEvent.IsValid())
{
delete pMessage->m_hReplyEvent;
pMessage->m_hReplyEvent = nullptr;

[noahfalk]

👍

[jkoritzinsky]

/ba-g WASM failure unrelated