[cDAC] Enumerate R2R hashmap pages into triage dumps for odd entry points

[max-charlamb]

Summary

Fix the cDAC's R2R stack walk aborting a frame past a SoftwareExceptionFrame when reading macOS triage minidumps. The root cause is an odd-entry-point optimization in the legacy DAC that skips a hashmap probe, so the bucket pages that probe would touch are never enumerated into the triage dump -- and the cDAC consumer then faults reading those not-in-dump pages.

Tracking issue: dotnet/diagnostics#5910 -- full repro, captured failure trace, and dump-file page-map evidence are there.

Root cause

ReadyToRunInfo::GetMethodDescForEntryPointInNativeImage looks up an entry point in the EntryPointToMethodDescMap hashmap. It has an x64/x86 fast path that returns early when the entry point is odd:

#if defined(TARGET_AMD64) || defined(TARGET_X86)
    // A normal method entry point is always 8 byte aligned, but a funclet can start at an odd address.
    if ((entryPoint & 0x1) != 0)
        return NULL;
#endif

This is purely a performance optimization: the map only ever contains true method entry points (4-byte aligned on every architecture), so a lookup for an odd (funclet) address is always a miss. (PtrHashMap handles odd keys fine -- they're just never present.)

The problem is the side effect in DAC builds. When createdump enumerates memory for a triage minidump (DOTNET_DbgMiniDumpType=3), it drives stack walking through this same path. The fast path makes it return without ever probing the hashmap for odd (funclet) entry points, so the bucket pages that probe would touch are never read and never captured into the dump.

The cDAC consumer, walking the same stack later, probes the hashmap for that odd entry point, lands in a bucket page absent from the dump, throws VirtualReadException, and aborts the walk -- SOS reports <failed> one frame past the SoftwareExceptionFrame.

Fix

Make the producer enumerate the pages instead of mirroring the bail on the consumer:

• src/coreclr/vm/readytoruninfo.cpp: gate the odd-entry-point optimization with !defined(DACCESS_COMPILE). The live runtime keeps the perf win; DAC builds perform the lookup so the bucket pages are enumerated into triage minidumps. Also removed the inaccurate "PtrHashMap can't handle odd pointers" comment.
• cDAC ExecutionManagerCore.ReadyToRunJitManager: removed the bail entirely. With the DAC enumerating the pages, the consumer reads them naturally.

This is a root-cause fix (complete dump enumeration) rather than masking the symptom on the consumer side, and it benefits the legacy DAC consumer too, not just the cDAC.

Validation

• Native build: rebuilt coreclr.dll (non-DAC) and mscordaccore.dll (DAC); both compilations of readytoruninfo.cpp succeed with 0 warnings / 0 errors, confirming the #if change is valid in both configurations.
• cDAC unit tests (Microsoft.Diagnostics.DataContractReader.Tests): all green.
• End to end: the failing diagnostics CI tests create fresh triage dumps, so once this change rides in via the runtime -> diagnostics flow, the dumps will contain the previously-missing bucket pages and the cDAC walk completes. (Dumps captured by an older, still-bailing DAC won't contain those pages; this fix applies to newly created dumps.)

Note

This PR description was drafted with assistance from GitHub Copilot.

[Copilot]

Pull request overview

This PR updates the cDAC ReadyToRun MethodDesc lookup to short-circuit on x86/x64 when the computed entry point is odd (low bit set), returning TargetPointer.Null rather than probing EntryPointToMethodDescMap. This aligns the cDAC behavior with the legacy DAC’s fast-path and avoids doing an unnecessary hash map lookup for entry points that won’t be present as keys.

Changes:

• Add an architecture-gated odd-entry-point bail-out in GetMethodDescForRuntimeFunction (x86/x64 only).
• Minor formatting-only adjustment in the ReadyToRun major-version → GCInfo-version switch expression (no behavioral change).

[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

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 2 changed files in this pull request and generated 2 comments.