[release/6.0] JIT: fix bug where a gc struct is not zero initialized

[AndyAyersMS]

Port of #67825 to release/6.0.

Customer Impact

Reported by RavenDB (see #65694).

This bug can lead to unexpected crashes during GC.

Details

An optimization added in .NET 5 can in some cases lead to the jit reporting an uninitialized struct field as a gc reference.

The attached test case shows a simple app that exhibits this problem. The key components are:

• a struct type S with one or more ref-class fields
• a method that explicitly initializes a local s of type S,
  • then updates s via Dictionary.TryGetValue using the .? operator for the dictionary reference,
  • then only accesses the fields of s ifTryGetValue returns true

// "gc" structstructS{objecto;}publicintG(Keyk,Dictionary<Key,S>d){Ss=default;// explicit initializationif(d?.TryGetValue(k,outs)==true&&(s.o!=null)

Testing

Verified with a local RavenDB repro and new test case.

Risk

Low. The IL pattern that leads to this situation is not common. Fix had very minor diffs in SPMI.

[AndyAyersMS]

@BruceForstall PTAL (manual merge)

cc @dotnet/jit-contrib

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Port of #67825 to release/6.0.

Customer Impact

Reported by RavenDB (see #65694)

Details

An optimization added in .NET 5 can in some cases lead to the jit reporting an uninitialized struct field as a gc reference.

Testing

Verified with a local RavenDB repro and new test case.

Risk

Low. The IL pattern that leads to this situation is not common. Fix had very minor diffs in SPMI.

Author:	AndyAyersMS
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[jeffschwMSFT]

Approved. We will take this for consideration in 6.0.x