[release/6.0] Restore FP pairs when unwinding in ARM64

[hoyosjs]

Description

This change ensures that unwinding codes of the type save_fregp that are emitted for sequences like stp dX, dY, [sp, #Z] are correctly unwound. Without this, non-volatile registers are not correctly restored and frames above the handler frames can experience data corruption in the case of an exception in Unix based systems.

Customer Impact

Corruptions caused by this bug are hard to diagnose and lead to silent data loss.

Regression

No.

Testing

Added regression test that fails without the change.

Risk

Very low - equivalent fix exists in the OS tree.

Package authoring signed off?

N/A

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[jeffschwMSFT]

Approved. We are considering for 6.0.x

[jeffschwMSFT]

@hoyosjs please ensure that this has a code review.

[davidwrighton]

This is a good targetted fix for .NET 6.0

[hoyosjs]

This test is failing even against 6.0.5 on lab machines for Windows (basically without this change. This code path doesn't get used on Windows, which means the OS unwinder in the lab machines is likely not patched? At least for DDARM64-173, although that one reports having (Windows 10.0.19044.1645). I've seen it also fail in 10.0.19041.1415. This passes on windows 11 though. I could turn the test off for Windows given that it's only testing the OS at that point, but feels odd to be in that spot.

[mmitche]

Reminder that code complete is today.

[davidwrighton]

@hoyosjs I've confirmed that the OS unwinder does not have this fix on that version of Windows. Please just turn the test off for Windows, and we'll figure out what to do to best handle the situation later.

[jeffschwMSFT]

@hoyosjs are we still investigating PR failures, or are we ready to merge?

cc @carlossanlop