Porting an assembler, debugger, and more to WebAssembly November 30, 2014 on Drew DeVault's blog

WebAssembly is pretty cool! It lets you write portable C and cross-compile it to JavaScript so it’ll run in a web browser. As the maintainer of KnightOS, I looked to WASM as a potential means of reducing the cost of entry for new developers hoping to target the OS.

Rationale for WASM

There are several pieces of software in the toolchain that are required to write and test software for KnightOS:

You also need a copy of the latest kernel and any of your dependencies from packages.knightos.org. Getting all of this is not straightforward. On Linux and Mac, there are no official packages for any of these tools. On Windows, there are still no official packages, and you have to use Cygwin on top of that. The first step to writing KnightOS programs is to manually compile and install several tools, which is a lot to ask of someone who just wants to experiment.

All of the tools in our toolchain are written in C. We saw WASM as an opportunity to reduce all of this effort into simply firing up your web browser. It works, too! Here’s what was involved.

Note: Click the screen on the emulator to the left to give it your keyboard. Click away to take it back. You can use your arrow keys, F1-F5, enter, and escape (as MODE).

The final product

Let’s start by showing you what we’ve accomplished. It’s now possible for curious developers to try out KnightOS programming in their web browser. Of course, they still have to do it in assembly, but we’re working on that 😉. Here’s a “hello world” you can run in your web browser:

We can also install new dependencies on the fly and use them in our programs. Here’s another program that draws the “hello world” message in a window. You should install core/corelib first:

You can find more packages to try out on packages.knightos.org. Here’s another example, this one launches the file manager. You’ll have to install a few packages for it to work:

Install:

Feel free to edit any of these examples! You can run them again with the Run button. These resources might be useful if you want to play with this some more:

z80 instruction set - z80 assembly tutorial - KnightOS reference documentation

Note: our toolchain has some memory leaks, so eventually WASM is going to run out of memory and then you’ll have to refresh. Sorry!

How all of the pieces fit together

When you loaded this page, a bunch of things happened. First, the latest release of the KnightOS kernel was downloaded. Then all of the WASM ports of the toolchain were downloaded and loaded. Some virtual filesystems were set up, and two KnightOS packages were downloaded and installed: core/init, and core/kernel-headers, respectively necessary for booting the system and compiling code against the kernel API. Extracting those packages involves copying them into kpack’s virtual filesystem and running kpack -e path/to/package root/.

When you click “Run” on one of these text boxes, the contents of the text box is written to /main.asm in the assembler’s virtual filesystem. The package installation process extracts headers to /include/, and scas itself is run with /main.asm -I/include -o /executable, which assembles the program and writes the output to /executable.

Then we copy the executable into the genkfs filesystem (this is the tool that generates filesystem images). We also copy the empty kernel into this filesystem, as well as any of the packages we’ve installed. We then run genkfs /kernel.rom /root, which creates a filesystem image from /root and bakes it into kernel.rom. This produces a ready-to-emulate ROM image that we can load into the z80e emulator on the left.

The WASM details

Porting all this stuff to WASM wasn’t straightforward. The easiest part was cross-compiling all of them to JavaScript:

cd build
emconfigure cmake ..
emmake make

The process was basically that simple for each piece of software. There were a few changes made to some of the tools to fix a few problems. The hard part came when I wanted to run all of them on the same page. WASM compiled code assumes that it will be the only WASM module on the page at any given time, so this was a bit challenging and involved editing the generated JS.

The first thing I did was wrap all of the modules in isolated AMD loaders1. You can see how some of this ended up looking by visiting the actual scripts (warning, big files):

That was enough to make it so that they could all run. These are part of a toolchain, though, so somehow they needed to share files. Emscripten’s FS object cannot be shared between modules, so the solution was to write a little JS:

copy_between_systems = (fs1, fs2, from, to, encoding) ->
  for f in fs1.readdir(from)
    continue if f in ['.', '..']
    fs1p = from + '/' + f
    fs2p = to + '/' + f
    s = fs1.stat(fs1p)
    log("Writing #{fs1p} to #{fs2p}")
    if fs1.isDir(s.mode)
      try
        fs2.mkdir(fs2p)
      catch
        # pass
      copy_between_systems(fs1, fs2, fs1p, fs2p, encoding)
    else
      fs2.writeFile(fs2p, fs1.readFile(fs1p, { encoding: encoding }), { encoding: encoding })

With this, we can extract packages in the kpack filesystem and copy them to the genkfs filesystem:

install_package = (repo, name, callback) ->
  full_name = repo + '/' + name
  log("Downloading " + full_name)
  xhr = new XMLHttpRequest()
  xhr.open('GET', "https://packages.knightos.org/" + full_name + "/download")
  xhr.responseType = 'arraybuffer'
  xhr.onload = () ->
    log("Installing " + full_name)
    file_name = '/packages/' + repo + '-' + name + '.pkg'
    data = new Uint8Array(xhr.response)
    toolchain.kpack.FS.writeFile(file_name, data, { encoding: 'binary' })
    toolchain.kpack.Module.callMain(['-e', file_name, '/pkgroot'])
    copy_between_systems(toolchain.kpack.FS, toolchain.scas.FS, "/pkgroot/include", "/include", "utf8")
    copy_between_systems(toolchain.kpack.FS, toolchain.genkfs.FS, "/pkgroot", "/root", "binary")
    log("Package installed.")
    callback() if callback?
  xhr.send()

And this puts all the pieces in place for us to actually pass an assembly file through our toolchain:

run_project = (main) ->
  # Assemble
  window.toolchain.scas.FS.writeFile('/main.asm', main)
  log("Calling assembler...")
  ret = window.toolchain.scas.Module.callMain(['/main.asm', '-I/include/', '-o', 'executable'])
  return ret if ret != 0
  log("Assembly done!")
  # Build filesystem
  executable = window.toolchain.scas.FS.readFile("/executable", { encoding: 'binary' })
  window.toolchain.genkfs.FS.writeFile("/root/bin/executable", executable, { encoding: 'binary' })
  window.toolchain.genkfs.FS.writeFile("/root/etc/inittab", "/bin/executable")
  window.toolchain.genkfs.FS.writeFile("/kernel.rom", new Uint8Array(toolchain.kernel_rom), { encoding: 'binary' })
  window.toolchain.genkfs.Module.callMain(["/kernel.rom", "/root"])
  rom = window.toolchain.genkfs.FS.readFile("/kernel.rom", { encoding: 'binary' })

  log("Loading your program into the emulator!")
  if current_emulator != null
    current_emulator.cleanup()
  current_emulator = new toolchain.ide_emu(document.getElementById('screen'))
  current_emulator.load_rom(rom.buffer)
  return 0

This was fairly easy to put together once we got all the tools to cooperate. After all, these are all command-line tools. Invoking them is as simple as calling main and then fiddling with the files that come out. Porting z80e, on the other hand, was not nearly as simple.

Porting z80e to the browser

z80e is our calculator emulator. It’s also written in C, but needs to interact much more closely with the user. We need to be able to render the display to a canvas, and to receive input from the user. This isn’t nearly as simple as just calling main and playing with some files.

To accomplish this, we’ve put together OpenTI, a set of JavaScript bindings to z80e. This is mostly the work of my friend puckipedia, but I can explain a bit of what is involved. The short of it is that we needed to map native structs to JavaScript objects and pass JavaScript code as function pointers to z80e’s hooks. So far as I know, the KnightOS team is the only group to have attempted something with this deep of integration between WASM and JavaScript - because we had to do a ton of the work ourselves.

OpenTI contains a wrap module that is capable of wrapping structs and pointers in JavaScript objects. This is a tedious procedure, because we have to know the offset and size of each field in native code. An example of a wrapped object is given here:

define(["../wrap"], function(Wrap) {
  var Registers = function(pointer) {
    if (!pointer) {
      throw "This object can only be instantiated with a memory region predefined!";
    }
    this.pointer = pointer;

    Wrap.UInt16(this, "AF", pointer);
    Wrap.UInt8(this, "F", pointer);
    Wrap.UInt8(this, "A", pointer + 1);

    this.flags = {};
    Wrap.UInt8(this.flags, "C",  pointer, 128, 7);
    Wrap.UInt8(this.flags, "N",  pointer,  64, 6);
    Wrap.UInt8(this.flags, "PV", pointer,  32, 5);
    Wrap.UInt8(this.flags, "3",  pointer,  16, 4);
    Wrap.UInt8(this.flags, "H",  pointer,   8, 3);
    Wrap.UInt8(this.flags, "5",  pointer,   4, 2);
    Wrap.UInt8(this.flags, "Z",  pointer,   2, 1);
    Wrap.UInt8(this.flags, "S",  pointer,   1, 0);
    pointer += 2;

    Wrap.UInt16(this, "BC", pointer);
    Wrap.UInt8(this, "C", pointer);
    Wrap.UInt8(this, "B", pointer + 1);
    pointer += 2;

    Wrap.UInt16(this, "DE", pointer);
    Wrap.UInt8(this, "E", pointer);
    Wrap.UInt8(this, "D", pointer + 1);
    pointer += 2;

    Wrap.UInt16(this, "HL", pointer);
    Wrap.UInt8(this, "L", pointer);
    Wrap.UInt8(this, "H", pointer + 1);
    pointer += 2;

    Wrap.UInt16(this, "_AF", pointer);
    Wrap.UInt16(this, "_BC", pointer + 2);
    Wrap.UInt16(this, "_DE", pointer + 4);
    Wrap.UInt16(this, "_HL", pointer + 6);
    pointer += 8;

    Wrap.UInt16(this, "PC", pointer);
    Wrap.UInt16(this, "SP", pointer + 2);
    pointer += 4;

    Wrap.UInt16(this, "IX", pointer);
    Wrap.UInt8(this, "IXL", pointer);
    Wrap.UInt8(this, "IXH", pointer + 1);
    pointer += 2;

    Wrap.UInt16(this, "IY", pointer);
    Wrap.UInt8(this, "IYL", pointer);
    Wrap.UInt8(this, "IYH", pointer + 1);
    pointer += 2;

    Wrap.UInt8(this, "I", pointer++);
    Wrap.UInt8(this, "R", pointer++);

    // 2 dummy bytes needed for 4-byte alignment
  }

  Registers.sizeOf = function() {
    return 26;
  }

  return Registers;
});

The result of that effort is that you can find out what the current value of a register is from some nice clean JavaScript: asic.cpu.registers.PC (it’s , by the way). Pop open your JavaScript console and play around with the current_asic global!

Conclusions

I’ve put all of this together on try.knightos.org. The source is available on GitHub. It’s entirely client-side, so it can be hosted on GitHub Pages. I’m hopeful that this will make it easier for people to get interested in KnightOS development, but it’ll be a lot better once I can get more documentation and tutorials written. It’d be pretty cool if we could have interactive tutorials like this!

If you, reader, are interested in working on some pretty cool shit, there’s a place for you! We have things to do in Assembly, C, JavaScript, Python, and a handful of other things. Maybe you have a knack for design and want to help improve it. Whatever the case may be, if you have interest in this stuff, come hang out with us on IRC: #knightos on irc.freenode.net.


2018-08-31: This article was updated to fix some long-broken scripts and adjust everything to fit into the since-updated blog theme. The title was also changed from “Porting an entire desktop toolchain to the browser with Emscripten” and some minor editorial corrections were made. References to Emscripten were replaced with WebAssembly - WASM is the standard API that browsers have implemented to replace asm.js, and the Emscripten toolchain and JavaScript API remained compatible throughout the process.


  1. AMD was an early means of using modules with JavaScript, which was popular at the time this article was written (2014). Today, a different form of modules has become part of the JavaScript language standard. ↩︎

Articles from blogs I read Generated by openring

Robust generic functions on slices

Avoiding memory leaks in the slices package.

via The Go Blog February 22, 2024

ClassNotFoundException: java.util.SequencedCollection

Recently I’ve had users of my libraries start reporting mysterious errors due to a missing reference to SequencedCollection, a Java interface added in JDK 21: Execution error (ClassNotFoundException) at jdk.internal.loader.BuiltinClassLoader/loadClass (Buil…

via Aphyr: Posts February 20, 2024

Status update, February 2024

Hi! February is FOSDEM month, and as usual I’ve come to Brussels to meet with a lot of other FOSS developers and exchange ideas. I like to navigate between the buildings and along the hallways to find nice people to discuss with. This edition I’ve been invol…

via emersion February 20, 2024