triggering SNES revision 1 CPU DMA/HDMA crash?

[lidnariq]

Does anyone know how to reliably trigger the DMA/HDMA crash? I was hoping that we might be able to figure out what was going on using poot36's slightly-broken SNES, but byuu tells me it's not simply as easy as just starting one of each transfer type at the same time.

[Ramsis]

I struggled with this very bug in an early alpha build of my PowerPak firmware. I could try and write up a test ROM that reproduces it (the original code unfortunately wasn't preserved, but I do remember the details). This might take a few days though as I have little spare time right now ...

[lidnariq]

If you have any down time, I'd appreciate it.

Only caveat is that poot36's SNES breaks the stack pointer when you issue a PLB or PLD instruction, but you probably don't need those.

[Ramsis]

Here you go.

dma-hdma.png

Source code is included so you can tinker around with it if you feel like it. BTW, I ensured there were no plb/pld instructions used.

[Ramsis]

Screenshots of some of the less dramatic effects:







I especially like the solid green background color it brings up at some point. Note how the text cursor position gets messed up as well.

[mikejmoffitt]

This is beautiful and weird; is the cause of this crash written up somewhere? A little searching didn't return much.

[lidnariq]

I was hoping to trigger the crash on poot36's SNES (since he has a logic analyzer there) but it hasn't been cooperative yet.

[Near]

> This is beautiful and weird; is the cause of this crash written up somewhere? A little searching didn't return much.

Nobody knows.

Given that it crashes the program execution, I have no way to write a software test to verify what's going on.

My best guess is that it ends up fetching (H)DMA data for instructions to execute and/or fetching instructions [and subsequently displacing the program counter] for (H)DMA to process.

But from over a decade of past experience, it'll most likely end up being far more convoluted than that. When it comes to the SNES, it's never the most logical behavior.

[tepples]

I'm guessing an address bus conflict of some sort between the CPU and the HDMA controller.