Add support for no GC regions in x86 GC info

[filipnavara]

Contributes to #113985

#113985 (comment) identified an issue where GC info is not correctly recorded for inline jumps to finally blocks. It shows up in GC stress testing. This issue already exists on NativeAOT on win-x86 but we didn't have sufficient coverage to identify it.

For fully interruptible methods we can now encode a table in the x86 GC info that describes the no-GC regions as pair of (offset, size) tuples. The presence of the table is indicated by a new header opcode (SET_NOGCREGIONS_CNT + count for count >= 0 && count <= 4 or FFFF_NOGCREGION_CNT and extra count field after the header).

[filipnavara]

cc @dotnet/dotnet-diag If this gets approved the shared files for GC info parsing and dumping will need to be synced to the dotnet/diagnostics repo.

[filipnavara]

^ This is the primary issue addressed by this PR

[jkotas]

/azp run runtime-coreclr gcstress0x3-gcstress0xc

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[filipnavara]

/azp run runtime-coreclr gcstress0x3-gcstress0xc

I don't expect this to reveal anything. It's now a strictly zero-diff change with x86 funclets disabled. May be worth running the same check on #113576 which has essentially the same commit applied with x86 funclets enabled (- the disabled cpObj optimization).

(In fact, give me a sec, I'll push the exact same change there... and DONE.)

[jkotas]

I don't expect this to reveal anything.

Right, I just want to make sure that there is nothing subtle broken in the x86 GC encoder/decoder.

[davidwrighton]

I've taken a pass through this and it looks good, although I haven't looked at the test failures to see if they are related.

[filipnavara]

I haven't looked at the test failures to see if they are related.

The osx failure is unrelated. I will check the GC stress failures tomorrow. I believe I have previously seen them and there's one issue filed for x86 unwinding (broken by async2 codegen change using a slightly different pattern in few epilogs).

[filipnavara]

The two x86 GC stress failures are indeed instances of #115120.

[jkotas]

Thank you

[jkotas]

/ba-g intermittent nuget restore failure in unrelated leg

[jkotas]

If this gets approved the shared files for GC info parsing and dumping will need to be synced to the dotnet/diagnostics repo.

Could you please send matching update PR to the diagnostics repo?

[filipnavara]

Could you please send matching update PR to the diagnostics repo?

Of course.