Page 1 of 1

Coredump tool

Posted: Mon Aug 11, 2014 8:37 pm
by JRoatch
Screen shot of $0400-$047f after booting this from a powerpak.
Screen shot of $0400-$047f after booting this from a powerpak.
coredump-v1.0-screenshot.jpg (37.04 KiB) Viewed 11539 times
Coredump is a tool that prints the contents of the NES internal ram given a key combo on reset. This is an improved version of the stack overflow screen I did awhile back. When installed at the reset vector this can be use for two things: for debugging games that crash while on a unmodified console, and for scientifically getting data on the initial contents of NES ram. Perhaps even convincing emulator authors to ditch the arbitrary 0x00000000ffffffff pattern.

I hope that even though this compiles to a hefty 485 bytes, that it can be part of NES power-on self-test (POST)

Re: Coredump tool

Posted: Tue Aug 12, 2014 1:10 am
by thefox
43110 wrote:Perhaps even convincing emulator authors to ditch the arbitrary 0x00000000ffffffff pattern.
I think that would be only FCEUX, though. :)

The test program is nice and simple, I like it!

Re: Coredump tool

Posted: Wed Aug 13, 2014 2:07 am
by Myask
Myask wrote:I wonder how much NES state the PowerPak clobbers between boot and game.
Perhaps you should write a DEADBEEF or other RAM filler, to run before this, and see if RAM is part of it.

Obviously the PowerPak hits PPU registers, CPU status bits, and one register to LD* ST* in order to hit the PPU registers. It must also discard whatever data were in the controllers' shift registers, as it must to read from any of them.
---
[semi-offtopic hunk split off to start another topic]

Re: Coredump tool

Posted: Wed Aug 13, 2014 12:15 pm
by JRoatch
The split post is relevant.

So I was working on this a bit (misfeature fixes, view nametables before overwriting, and sounds), and you got me thinking. Is it possible to stash the cpu registers by writing to OAM_DATA on reset, and retrieving them two blank frames later by cycling around OAM also with OAM_DATA? I don't plan on preserving OAM since it'll most likely have a mirror in RAM.

Edit: To answer my own question. No. OAM_DATA does not work like I expected. The best I can hope for is a brk with an IRQ vector that pushes A X and Y.

Re: Coredump tool

Posted: Thu Aug 14, 2014 3:53 pm
by JRoatch
Version 1.1 attached.
- $01 and stack pointer are now preserved in idle frames.
- added sounds.

The next version will be very different as I'll have to start using the nameables to store operating state to include the features I'm planning.

Re: Coredump tool

Posted: Fri Aug 22, 2014 3:42 pm
by JRoatch
Screen shot of part of the nametable after booting this from a powerpak.
Screen shot of part of the nametable after booting this from a powerpak.
coredump-v1.2-screenshot.jpg (39.85 KiB) Viewed 11304 times
Version 1.2 attached.
- This version is separated into a two parts: a boot backend and a GUI frontend. Boot stuffs things into CHR RAM, and GUI reads and prints them to the screen.
- New things now captured at boot are nametables and palette.

CHR RAM was decided to be used as scratch memory since everywhere else (except sram) is internal nes memory.

This will probably be the last version for a long while. If there is a next version, it will most likely be marked 2.0 along with a redesigned GUI.

Sorry if the multiple posts are getting annoying.

Re: Coredump tool

Posted: Mon Oct 26, 2015 6:26 am
by JRoatch
As I was chipping away at my framework code (and as a response to this)
I cleaned up the code a bit, and improved modularity.
The source is embedded in the rom, and will also come with an update to my framework code later.

Re: Coredump tool

Posted: Wed Feb 03, 2016 1:31 pm
by alphamule
Hmm, could a clever use of serial outputs avoid using even CHR-RAM?

Re: Coredump tool

Posted: Wed Feb 03, 2016 8:20 pm
by JRoatch
Probably.
Every version of coredump except v1.2 could technically use CHR-ROM, but the stack pointer needs to be restored from somewhere like the nametable because I use the stack pointer able to save byte $01.

There's also an issue of code size with simply using the same approach coredump does. Since one of the original goals was to preserve RAM, we can't use subroutines. So everything has to be duplicated inlined. While copies of the nibble printing code is OK, I wouldn't want to duplicate the multi-layered loops of bit banging serial output.