I have stepped through this in debugger. I see two problems:

1. corert/src/Native/Runtime/coreclr/gcinfodecoder.cpp

   Line 286 in 2b686a5

   m_ReversePInvokeFrameStackSlot = (INT32)m_Reader.DecodeVarLengthSigned(REVERSE_PINVOKE_FRAME_ENCBASE);

   is missing call to DENORMALIZE_STACK_SLOT similar to all other slot properties. Once you fix this problem, the slot should be right for the "GetSP()" codepath.

2. The implementation of CodeManager::GetFramePointer does not agree with where the JIT decides to use FP as the base. Here is small repro:

• Cherry pick https://github.com/jkotas/corert/commit/6cf515cffa1407085bf52b932b956771f0fd5ec7
• Compile and set breakpoint at RhpReversePInvoke2 and in this place in the code manager. The pFrame argument for RhpReversePInvoke2 should have same value as what code manager computes later.

To fix this problem, the base should be picked based on gcInfoDecoder.GetStackBaseRegister() instead.

It works significantly better after last fixes, but the test program finishes with 'ASSERT_UNCONDITIONALLY', when trying to "Reset nesting levels for exceptions on this thread that might not be currently in flight". It looks like there is an another issue, but this part works right.

the test program finishes with 'ASSERT_UNCONDITIONALLY', when trying to "Reset nesting levels for exceptions on this thread that might not be currently in flight

I do not see this problem. I agree that it is a different issue. Could you please open an issue with repro step details?