| x6502 |
| a simple 6502 CPU emulator |
x6502 is an emulator for the 6502 class of processors.
It currently supports the full instruction set of the
6502 (plus a few extensions) and has a rudimentary
simulated I/O bus. It should be able to run arbitrary
x6502 bytecode with ``correct'' results, although most
binaries for common 6502 systems (Atari, C64, Apple II,
etc) won't function as expected, since they expect I/O
devices to be mapped into memory where there are
x6502 is freely available under the original 4-clause
BSD license, the full text of which is included in the
Building and running x6502
To build x6502, just run `make' in the project root. You
will need clang and Python installed. To use gcc, change
the $(CC) var in the Makefile. No libraries beyond POSIX
libc are used. This will produce the x6502 binary.
x6502 takes the compiled 6502 object file as an
argument, and runs it until it encounters an EXT
instruction (EXT instructions are an extension to 6502
bytecode, see below). You can use any 6502 assembler to
compile to 6502 bytecode; `xa' is one that is bundled
with Debian-based distros. Note that, by default, x6502
loads code in at address 0x1000; you therefore need to
either tell your assembler that that's the base address
for the text section of your binary or override the
default load address using the `-b' flag of x6502. Note
that 0x1000 is the default load address for the `xa'
If you want to compile a version of x6502 that dumps
machine state after every instruction, run `make debug'
instead of `make'. This will also disable compiler
optimizations. This mode really does not play nice with
vterm mode, so be warned.
Extensions to the 6502 instruction set
x6502 recognizes two instructions that are not in the
original 6502 instruction set. These are:
DEBUG (0xFC): prints debugging information about the
current state of the emulator
EXT (0xFF): stops the emulator and exits
Both of these are defined as macros in `stdlib/stdio.s'.
To disable these extensions, compile with
-DDISABLE_EXTENSIONS (right now, this can be done by
adding that flag to the Makefile).
This also implements a subset of the 65C02 and 65C816
instruction set, in particular the WAI (0xCB)
instruction. The WAI instruction pauses the emulator
until an I/O interrupt is thrown.
I/O memory map:
There are four I/O devices right now: a character input
device, a character output device, a virtual terminal
and a block device. Convenience constants and macros for
the character I/O devices are defined in
`stdlib/stdio.s' for use in user programs. Add stdlib to
your include path and then add `#include <stdio.s>' to
your program to use these constants.
I/O options are controlled by setting bits on the I/O
flag byte at address 0xFF02. The current set of
supported flags are:
when set, activates vterm mode.
when set, waits for a keypress input before
terminating upon receiving an EXT instruction.
Non-vterm applications will probably want to set
this flag, as some implementations of ncurses
will clear the display when the emulator exits.
When outputting characters, you can control the
``paint'' with which the characters are drawn. You can
do so by modifying the PAINT flag at location 0xFEE8.
Paints are an OR-ing of a color (bottom 4 bits) and a
style (top 4 bits). Supported colors are:
Supported styles are:
Thus, as an example, an underlined, bold green character
would have paint 0xC2.
The character output device is mapped to 0xFF00. Any
character written to FF00 is immediately echoed to the
The character input device is mapped to 0xFF01. When a
character is available on standard in, an interrupt is
raised and FF01 is set to the character that was
received. Note that one character is delivered per
interrupt; if the user types ``abc'', they will get
three interrupts one after the other.
The virtual terminal is activated by setting the
VTERM_ENABLE bit on the IO flag byte. After the flag is
set, the data in memory addresses 0xFB00 through 0xFEE7
are mapped to a 40x25 grid in the host terminal. Data in
this region is stored in row-major format, and any write
will trigger a refresh of the vterm.
Note that even in vterm-mode, the putchar-esque
character output device is still usable, and will put
the character at the position directly after the
position of the last write.
A commented example of how to use the character I/O
capabilities of x6502 is provided in
sample_programs/echo.s, and an example of a vterm
application is provided in sample_programs/spam.s
A block device can be mapped in with control addresses
at 0xFF03 through 0xFF07. To use the block device, you
must specify a binary disk image to back the device
using the -d flag. To read from the block device, write
an address in the disk image to 0xFF03 and 0xFF04, with
the low byte in 0xFF03. The value at that location in
the disk image will be written to 0xFF05, which your
program can then read. To write, set the memory address
to write to using the same method, then write the
desired byte to 0xFF06. If any of these operations
return an error, the byte at 0xFF07 will be nonzero.
Reading the source
x6502 was written to be easy to understand and read. A
good place to start is `cpu.h', which defines a few
constants used throughout the code (mostly around CPU
flags) as well as the `cpu' struct, which is used pretty
`emu.c' is where the interesting stuff happens; this is
the main loop of the emulator where opcodes are decoded
as dispatched. It also handles interrupts and calls out
to I/O handlers.
The code for actual opcode interpretation is a little
strange; there are lots of ``header'' files in the
opcode_handlers directory that are not really header
files at all. These files all contain code for handling
opcode parsing and interpretation; with over 150
opcodes, having all of the code to handle these in one
file would be excessive and difficult to navigate, and
dispatching out to functions to handle each opcode
carries unnecessary overhead in what should be the
tightest loop in the project. Thus, each of these header
files is #included in emu.c in the middle of a switch
statement, and gets access to the local scope within the
main_loop function. It's weird but it gets the job done,
and is the least bad of all considered options.
The opcode handlers all use convenience functions
defined in `functions.h', most of which are for the
various addressing modes of the 6502 or for dealing with
`io.c' is where the I/O bus lives; this is where we
check to see if the emulated character device has been
written to and where we raise an interrupt if we've
gotten input from stdin.
`generate_debug_names.py' reads the `opcodes.h' header
and generates `debug-names.h', which contains a mapping
from opcode to a string representation of that opcode.
It's only used when dumping CPU state, either because
the DEBUG flag was set at compile time or because a
DEBUG instruction was hit in the binary.
The rest of the files are pretty boring; `main.c' is
pretty much only responsible for loading bytecode into
memory and parsing command line arguments and `debug.c' is
used to provide the `dump_cpu' function, which is a
fascinating function consisting of almost nothing but
- support buffered input, where the program can read
more than one input character at once.
- voltagex on Github for sending me a patch to improve
the sample_programs readme.
- anatoly on HN for suggesting I add a bit on source
code structure to the README.
- shalmanese for coffee and pie.
- daumiller for finding the subtraction bug.