Emulator should not trigger interrupts while skipping instructions.

DCPUTeam / DCPUToolchain

[ARCHIVED] The code repository for the DCPU-16 Toolchain.

http://dcputoolcha.in/

MIT License

97 stars 14 forks source link

Emulator should not trigger interrupts while skipping instructions. #224

Closed unbibium closed 11 years ago

unbibium commented 11 years ago

The DCPU-16 specification version 1.7 states: "Interrupts are not triggered while the DCPU-16 is skipping," implying that it should queue interrupts until the skip flag is clear. DTEMU does interrupt while sipping, causing unpredictable behavior.

I discovered this while writing a program that uses a clock interrupt, and I find that if the interrupt occurs immediately following a false IF opcode, then the first instruction of the interrupt is aso skipped. In the -d dump, it looks like this:

0xA280: IFE X NXT_LIT (0x0000)
0xB32E: (skipped) IFN A NXT (0xAEB0)

... then at the end of the routine:

0xAEE1: RFI A
0xA282: SET PC NXT_LIT (0xA236)
0xA237: SET PC POP

The instruction at 0xA282 should have been skipped, not the one at 0xB32E.

hyperfang8 commented 11 years ago

The same issue as mine https://github.com/DCPUTeam/DCPUToolchain/issues/223 I can't use dcputoolchain emulator because of this anymore :( A very serius bug!!

patflick commented 11 years ago

this will be merged into master soon, i accidentally fixed it in the new compiler branch :)

patflick commented 11 years ago

i'm sorry this took so long, I was really busy the last few days, and our other main devs have quit or taken a longer break from the toolchain.

patflick commented 11 years ago

@unbibium @hyperfang8 : can you confirm the fix please?

unbibium commented 11 years ago

I will, once I figure out how I built it in the first place -- it's complaining that it can't find SPHINX_EXECUTABLE and refusing to build anything, even though it says only the IDE requires it.

Is it part of the development build up on the main website now?

unbibium commented 11 years ago

OK; it looks like it works now.

patflick commented 11 years ago

reopening this issue, interrupts that get triggered from actual hardware (e.g. keyboard in vm_hw_keyboard.c) still have this issue.

This needs some redesign of our interrupt code. I will get on it

patflick commented 11 years ago

this is now properly fixed. no more issues with interrupts!

unbibium commented 11 years ago

neat. Is there a test for this? I'd like to try running it on the various other emulators and bother those authors too.

hach-que commented 11 years ago

We should construct some tests for this in our test suite. I'll prototype some tests this morning but as I don't have my PC it'll be impossible to actually run them.

unbibium commented 11 years ago

I recompiled off master yesterday, and still got the kind of random behavior I generally expect from interrupt problems. I think I'll put together a python filter that detects skipping problems and pipe dtemu through it.