Z80: Decompiler gets confused when mixing interrupt control and stack operations

[agatti]

Describe the bug
The decompiler creates an incorrect instruction flow. Maybe it gets confused by stack manipulation and interrupt control code being interleaved rather than nested.

To Reproduce
Steps to reproduce the behavior:

1. Load the attached test.bin file as a Z80 binary starting at $8000.
2. Let auto-analysis start - nothing will happen.
3. Start disassemble from the very first instruction, the decompiler window will fill up with some invalid code.

Expected behavior
In the decompiled code, enableMaskableInterrupts is supposed to appear after the write to DAT_io_00a1, not right after disableMaskableInterrupts.

Screenshots

Attachments

test.bin.gz

Environment (please complete the following information):

• OS: macOS 10.14.6
• Java Version: OpenJDK 11.0.5
• Ghidra Version: 9.1-DEV built from Ghidra_9.1_build tag

[emteere]

Weird, will try to reproduce.

[cbuschardt]

Did you remember to mark the io region as volatile in the "Window->Memory Map" window?

[agatti]

@cbuschardt Marking the IO region as volatile lets the decompiler work as intended, indeed. Thanks for the tip.

That said, is it still a bug in the decompiler? Values are written to the IO area, so there are fewer assumptions to be made from the data in question.