JIT: Fix another issue when optimizing out async calls late

[jakobbotsch]

genAsyncResumeInfoTable may be null if we optimize out all async calls, so handle this case too.

[Copilot]

Pull Request Overview

This PR fixes a potential null reference crash when async calls are optimized out late in the JIT compilation process. The issue occurs when compSuspensionPoints exists (created during early async transformation) but genAsyncResumeInfoTable remains null because all GT_RECORD_ASYNC_RESUME nodes were removed by late optimizations.

• Removes an assertion that incorrectly assumed genAsyncResumeInfoTable would always be non-null
• Adds null check before accessing genAsyncResumeInfoTable to prevent crashes
• Sets diagnostic native offsets to 0 when async resume info table doesn't exist

[jakobbotsch]

PTAL @dotnet/jit-contrib

[VSadov]

@jakobbotsch what are the typical symptoms of this bug? Is it a JIT-time crash or run time NRE?

[jakobbotsch]

@jakobbotsch what are the typical symptoms of this bug? Is it a JIT-time crash or run time NRE?

In checked builds you would see assert, and in release builds like JIT-time crashes.

I found this issue via Fuzzlyn. It's probably unlikely any of our tests are hitting it.

[VSadov]

I am looking at something that may be a Linux-specific GC hole. At least it has symptoms of a GC hole (a variable becomes null and causes NRE, but not always).

Sounds like it cannot be possibly related to this.

[jakobbotsch]

/ba-g Timeouts