<!doctype html> <html> <head> <meta charset="utf-8"> <title>Paying It Forward: Documenting your Hardware</title> <meta name="description" content="A framework for easily creating beautiful presentations using HTML"> <meta name="author" content="Sean "xobs" Cross"> <meta name="apple-mobile-web-app-capable" content="yes"> <meta name="apple-mobile-web-app-status-bar-style" content="black-translucent"> <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no"> <link rel="stylesheet" href="css/reveal.css"> <link rel="stylesheet" href="css/theme/lca2020.css" id="theme"> <!-- Theme used for syntax highlighting of code --> <link rel="stylesheet" href="lib/css/zenburn.css"> <!-- Printing and PDF exports --> <script> var link = document.createElement('link'); link.rel = 'stylesheet'; link.type = 'text/css'; link.href = window.location.search.match(/print-pdf/gi) ? 'css/print/pdf.css' : 'css/print/paper.css'; document.getElementsByTagName('head')[0].appendChild(link); </script> <!--[if lt IE 9]> <script src="lib/js/html5shiv.js"></script> <![endif]--> <style> /********************************************* * ZOOM REVERSE TRANSITION (i.e. zoom out) *********************************************/ .reveal .slides section[data-transition=zoomrev], .reveal.zoomrev .slides section:not([data-transition]) { transition-timing-function: ease; } .reveal .slides > section[data-transition=zoomrev].past, .reveal .slides > section[data-transition~=zoomrev-out].past, .reveal.zoomrev .slides > section:not([data-transition]).past { visibility: hidden; -webkit-transform: scale(0.2); transform: scale(0.2); } .reveal .slides > section[data-transition=zoomrev].future, .reveal .slides > section[data-transition~=zoomrev-in].future, .reveal.zoomrev .slides > section:not([data-transition]).future { visibility: hidden; -webkit-transform: scale(16); transform: scale(16); } .reveal .slides > section > section[data-transition=zoomrev].past, .reveal .slides > section > section[data-transition~=zoomrev-out].past, .reveal.zoomrev .slides > section > section:not([data-transition]).past { -webkit-transform: translate(0, 150%); transform: translate(0, 150%); } .reveal .slides > section > section[data-transition=zoomrev].future, .reveal .slides > section > section[data-transition~=zoomrev-in].future, .reveal.zoomrev .slides > section > section:not([data-transition]).future { -webkit-transform: translate(0, -150%); transform: translate(0, -150%); } </style> </head> <body> <!-- Start of main presentation --> <div class="reveal"> <div class="footer"> <a class="url" href="https://p.xobs.io/lca20-pif/">p.xobs.io/lca20-pif</a> <span class="theme">Who’s Watching</span><span class="hashtag"> | #LCA2020</span><span class="twitter"> | @linuxconfau</span> </div> <div class="slides"> <section data-background-image="css/theme/lca2019-title-bg-transparent.svg"> <h2 style="background-color: transparent;">Paying it Forward: Documenting Your Hardware Project</h2> <h5 style="background-color: transparent;">Approaches to documenting a hardware description language using lxsocdoc</h5> <p align="right"> <small>Sean Cross - <a href="https://xobs.io/">https://xobs.io/</a> - @xobs</small> </p> </section> <section> <h2>Undocumented Hardware = Bad</h2> <h4 class="fragment">(But so easy to do!)</h4> <aside class="notes"> Undocumented hardware is bad. There are all sorts of quirks, and even if you have the source code, it can be very difficult to read. I'm the primary developer for the Fomu project, and this talk will cover some of the issues I've run into with respect to documentation. It is most directly related to the LiteX and Migen projects, but the concepts will carry over into any other Hardware Description Language you may use. The goal of this talk is to show how it's easy to document hardware with the right framework, and how it's easier to have a project that's documented than one that isn't. </aside> </section> <section> <h2>About Me</h2> <table> <tr> <td><img data-src="img/me.jpg"></td> <td><img data-src="img/fomu.png"></td> </tr> </table> <aside class="notes"> My name is Sean Cross, also known as "xobs". I will be speaking later this week on the Betrusted project, but many know me as the main developer behind the Fomu project. Fomu is an FPGA that fits in your USB port. One of my goals with the Fomu project was to allow people to treat it as just a RISC-V CPU in their USB port, which means now we need to make documentation. This talk covers some of the problems I ran into while working on this project, and the solutions I came up with. </aside> </section> <section> <h2>Reference Manuals</h2> <img class="fragment" data-src="img/rm-example.png"> <h6 class="fragment">Datasheet ≠ Reference Manual</h6> <aside class="notes"> How many people here know what a reference manual is? These are documents that you hopefully get from the manufacutrer of a chip that tell you how to use it. These are the single most important part of developing a chip the normal way, as they give you all of the information on how to use a chip. They can include everything from which memory addresses to poke to cause a pin to become an output, or how to set up the video block to output data. Here is the first page from the reference manual for the i.MX6 used in the Novena. You'll note that "Chapter 1: Introduction" starts on page 197 -- everything before it was a table of contents -- and is almost 6000 pages long. As an aside, reference manuals aren't datasheets -- datasheets tend to include electrical and operating information, and tend to be much shorter. </aside> </section> <section> <h2>Enterprise Documentation</h2> <table> <tr> <td> <img height="400" class="fragment" data-src="img/Microsoft_Office_Word.svg" type="image/svg+xml"/> </td> <td width="50%"> <div class="fragment"> <img data-src="img/Git-logo.svg" type="image/svg+xml"/> <img data-src="img/Travis_CI_Logo.svg" type="img/svg+xml"/> </div> </td> </tr> </table> <aside class="notes"> So how is this monster of a document written? I've asked several documentation writers from professional chip vendors, and they say that they have a whole team of people using a very advanced piece of realtime typesetting software. Most enterprise documentation is written by a team of writers using Microsoft Word. We're open source developers, and we prefer to use our own tools such as Git and CI to create documentation for us. Because we're lazy. And prone to making mistakes. </aside> </section> <section> <h2>Talk Outline</h2> <ol> <li>How to write HDL Code</li> <li>Rationale behind <tt>lxsocdoc</tt></li> <li>Examples of <tt>lxsocdoc</tt></li> <li>Benefits of this approach</li> </ol> <aside class="notes"> I'll briefly cover various methods of writing HDL code, then cover the rationale behind the approach we take with lxsocdoc, then give an example of how to use lxsocdoc and how you might apply it to your language. Finally, I'll cover the implications of having documented hardware and how this will help you pay it forward. </aside> </section> <section> <h2>Motivation</h2> <pre><code class="hljs cpp">// Hardware definitions of the SoC. Also is the main repo of // documentation for the programmer-centric view // of the hardware. /* Start of memory range for the UART peripheral */ #define UART_OFFSET 0x10000000 /* Offset of the data register for the debug UART. A write here will send the data out of the UART. A write when a send is going on will halt the processor until the send is completed. A read will receive any byte that was received by the UART since the last read, or 0xFFFFFFFF when none was. There is no receive buffer, so it's possible to miss data if you don't poll frequently enough. The debug UART is always configured as 8N1. */ #define UART_DATA_REG 0x00</code></pre> <p><tt>mach_defines.h</tt>, Hackaday 2019 Con Badge</p> <aside class="notes"> Verilog and VHDL are kind of the C or assembly of the FPGA world. They're universal, but somewhat unwieldy to use. You need to manually set up your address decoders, and documentation is very free-form. Common approaches today involve comments in the HDL and/or C header files. This works, but we can do better. We just need to describe the hardware better. </aside> </section> <section> <h2>Lots of Documentation</h2> <img data-src="img/rm-page-numbers.png"> <aside class="notes"> This documentation is very extensive. The start of Chapter 1 is on page 197, with all previous pages being the Table of Contents. It's almost 6000 pages. It's very extensive, because it's a very complicated chip. </aside> </section> <section> <h2>About LiteX</h2> <ul> <li>Hardware Description Language embedded in Python</li> <ul> <li>Doesn't run Python in hardware!</li> </ul> <li>Emits Verilog (or Yosys netlists)</li> <li>Makes it easy to create a SoC</li> <li>Powers the LCA2020 video production setup</li> </ul> <aside class="notes"> Fomu uses LiteX, which is related to Migen. This is a hardware description language written in Python. You write Python code and run the program, and it generates a design file -- either Verilog code, or a Yosys netlist. There are many other alternatives such as SpinalHDL or Chisel. By writing in Python as opposed to direct Verilog, we get a lot of nice primitives. The examples from this talk are taken from lxsocdoc and LiteX, but most higher-level hardware description languages can take similar approaches to documentation. </aside> </section> <section> <h2>LiteX Primitives</h2> <pre><code class="python" data-trim>class GPIOOut(Module, AutoCSR): def __init__(self, signal): self._out = CSRStorage(len(signal)) self.comb += signal.eq(self._out.storage)</code></pre> <aside class="notes"> In LiteX, two of the primitives used to expose hardware registers to the CPU softcore are CSRStorage and CSRStatus. Instead of manually wiring up a crossbar and decoding the addresses ourselves, we just need to write `self.regname = CSRStatus(8)`, and the build system will wire up 8 bits of read-only memory to the target CPU. Similarly, `self.othername = CSRStorage(8)` will give 8-bits of write-only memory. </aside> </section> <section> <h4>Case Study: SPI Bitbang Module</h4> <pre><code class="python" data-trim>self.bitbang = CSRStorage(4) If(self.bitbang.storage[3], dq.oe.eq(0) ).Else( dq.oe.eq(1) ), # CPOL=0/CPHA=0 or CPOL=1/CPHA=1 only. If(self.bitbang.storage[1], self.miso.status.eq(dq.i[1]) ), dq.o.eq( Cat(self.bitbang.storage[0], Replicate(1, spi_width-1)) )</code></pre> <aside class="notes"> This works well, but exposes a new problem: Documentation. As an example, I was working with the SPI Flash block in litex, and wanted to know how the bitbang driver worked. There wasn't any documentation except the source, which looked like this. You can kind of understand it, but it does take a lot of mental power to work through it. I started by creating aliases for the various elements in the storage array, but then I thought: There has to be a better way! </aside> </section> <section> <h2>Aside: Python Docstrings</h2> <pre><code class="python" data-trim>def _format_cmd(cmd, spi_width): """ `cmd` is the read instruction. Since everything is transmitted on all dq lines (cmd, adr and data), extend/ interleave cmd to full pads.dq width even if dq1-dq3 are don't care during the command phase: For example, for N25Q128, 0xeb is the quad i/o fast read, and extended to 4 bits (dq1,dq2,dq3 high) is: 0xfffefeff """ c = 2**(8*spi_width)-1 for b in range(8): if not (cmd>>b)%2: c &= ~(1<<(b*spi_width)) return c</code></pre> <aside class="notes"> As an aside, Python has something called Pydoc and Docstrings. These are comments that go at the top of functions and classes that let you describe what a Python object is and how to use it. This is almost what we want, except once the final SoC is generated we don't really care so much about things like constructor arguments or method properties. Documentation for the end user is different from documentation for the module developer. </aside> </section> <section> <h2>New Register Definition</h2> <pre><code class="python" data-trim>self.bitbang = CSRStorage(4, fields=[ CSRField("mosi", description="Output value for MOSI..." CSRField("clk", description="Output value for SPI CLK..." CSRField("cs_n", description="Output value for SPI C..." CSRField("dir", description="Sets the dir...", values=[ ("0", "OUT", "SPI pins are all output"), ("1", "IN", "SPI pins are all input"), ]) ], description="""Bitbang controls for SPI output. Only standard 1x SPI is supported, and as a result all four wires are ganged together. This means that it is only possible to perform half-duplex operations, using this SPI core.""")</code></pre> <aside class="notes"> This is when I hit upon the idea of `lxsocdoc`. The basic idea is that Python is really good at introspecting Python, so let's add a little bit more information to the CSR objects to make our life easier. And so, after working with the LiteX creator Florent, we refactored the bitbang definition to this. </aside> </section> <section> <h2>Refactored SPI Bitbang</h2> <pre><code class="python" data-trim>If(self.bitbang.fields.dir, dq.oe.eq(0) ).Else( dq.oe.eq(1) ), # CPOL=0/CPHA=0 or CPOL=1/CPHA=1 only. If(self.bitbang.fields.clk, self.miso.status.eq(dq.i[1]) ), dq.o.eq( Cat(self.bitbang.fields.mosi, Replicate(1, spi_width-1)) )</code></pre> <aside class="notes"> Now the actual bitbang logic looks like this. This is a little bit easier to understand -- no longer are we looking at indices in an array to determine what field does what. Instead we get actual named fields. But because Python can introspect Python very easily, this is just the beginning. </aside> </section> <section> <h2>Generating a Manual</h2> <img data-src="img/lxspi_bitbang.png"> <aside class="notes"> After the design is elaborated and the output file is generated, we can iterate through the resulting tree and pick out any CSR objects and using any additional information. We can actually generate a full reference manual, just like one you would receive from a SoC Vendor. For example, this is what the start of the Fomu SPI Flash documentation looks like: [Register Listing for LXSPI] This is already pretty useful. You can hand this file to someone and show them how your design works. But the title of this talk is called "Paying it Forward", and I can tell you from experience that having such a reference manual for yourself while developing software for your own hardware is still invaluable. Hardware designs are complex things, and not having to decode bitfield offsets in your head or constantly referring to various sections of code to see how it's implemented saves valuable time. </aside> </section> <section> <h2>Interrupts</h2> <img data-src="img/interrupts.png"> <aside class="notes"> We can even extract interrupt information, including which bits inside an interrupt register map to which event, and which interrupt number is assigned to a given module. </aside> </section> <section> <h2>Undocumented Fields</h2> <img data-src="img/timer0-event.png"> <aside class="notes"> It turns out that there is enough information that we can extract that even undocumented fields are somewhat useful. This is an undocumented interrupt register, but lxsocdoc has pulled out the field names and is giving useful documentation anyway. </aside> </section> <section> <h2>More Documentation: ModuleDoc</h2> <img data-src="img/timer0-doc.png"> <aside class="notes"> So now we have register documentation. Can we do better? Of course we can. SoC reference manuals are more than just register definitions. They also include background information on protocols, as well as more elaboration on how the block works. We can take a cue from CSRs themselves, and add module documentation in a similar fashion. </aside> </section> <section> <h2>Protocol Documentation</h2> <img data-src="img/usb-wishbone.png"> <aside class="notes"> We can add additional documentation such as protocol waveforms. Here we use WaveDrom to define the protocol of Wishbone-over-SPI. There are multiple formats of the protocol depending on which version is instantiated. </aside> </section> <section> <h2>SVD: Documentation for Machines</h2> <table> <tr> <td> <ul> <li>XML description file</li> <li>Interrupt numbers</li> <li>Memory layout</li> <li>Register definitions</li> <li>Register fields</li> </ul> </td> <td> <img data-src="img/CMSIS_SVD_Schema_Gen.png"> <br/><small>Source: keil.com</small> </td> </tr> </table> <aside class="notes"> Having documentation for humans is great, but we can go one step further and make documentation for computers. SVD is an XML format defined by ARM that defines various aspects about a chip, including memory layout, interrupt map, and register sets. SVD includes information such as default values and field bits, all information we have thanks to the introspectability of Python. </aside> </section> <section> <h2>SVD2Rust: Generating Safe Accessors</h2> <pre><code class="rust">fn init(&mut self) { self.registers .ctrl .write(|w| w.exe().bit(true).curren().bit(true).rgbleden().bit(true)); self.write(LEDDEN | FR250 | QUICK_STOP, LedRegister::LEDDCR0); // Set clock register to 12 MHz / 64 kHz - 1 self.write(((12_000_000u32 / 64_000u32) - 1) as u8, LedRegister::LEDDBR); self.write( BREATHE_ENABLE | BREATHE_MODE_FIXED | breathe_rate_ms(128), LedRegister::LEDDBCRR, ); }</code></pre> <aside class="notes"> In addition to generating a reference manual for humans, we can generate an SVD file that's usable in a wide variety of areas. For example, we can turn an SVD file into a Rust Peripheral Access Crate (PAC) using `SVD2Rust`, giving us an easy way to safely access all peripherals on a device. </aside> </section> <section> <h2>Renode: Fancy Register Logging</h2> <table width="100%"> <tr style="padding-right: 0px; padding-left: 0px;"> <td width="75%" style="padding-right: 0px; padding-left: 0px; text-align: center"> <img style="padding-right: 0px; padding-left: 0px;" width="100%" data-src="img/renode-debug.png"/> </td> <td style="padding-right: 0px; padding-left: 0px; text-align: center"> <img data-src="img/renode-ui-tall.png"> </td> </tr> </table> <aside class="notes"> We can also import this SVD file into an emulator such as Renode, which will print out fields and flags that get accessed, giving us greater visibility into what a program is doing. </aside> </section> <section> <h2>Benefits of Higher Level Languages</h2> <ul> <li>Greater code reuse</li> <li>More hardware description</li> <li>Better interoperability</li> <li>Automatic document generation</li> <li>Automatic SVD</li> </ul> <aside class="notes"> By using a higher level language, we are able to describe the hardware in greater detail than if we used Verilog or VHDL. We can add additional fields to our register definition fields to provide nice, human-readable documentation. This also allows us to generate machine-readable formats such as SVD, which opens up a whole world of software. </aside> </section> <section> <h2>Documentation helps others</h2> <h2 class="fragment">Documentation helps you</h2> <aside class="notes"> Documenting your hardware is important because it is necessary for you to write software that interfaces with it today, and it helps you work with others when it comes time to share your design with the world. By properly documenting various fields within your module, you make it easier on yourself to interact with today, and you make it easier to let others get up to speed in the future. By documenting your hardware, you're helping to pay it forward. </aside> </section> <section> <h2>Thank you</h2> <h3>Questions</h3> </section> </div> </div> <!-- class="reveal" --> <!-- End of main presentation --> <!-- Start of configuration section --> <script src="lib/js/head.min.js"></script> <script src="js/reveal.js"></script> <script> var presenter = !!Reveal.getQueryHash().s; // More info https://github.com/hakimel/reveal.js#configuration Reveal.initialize({ controls: presenter ? false : true, progress: true, history: true, center: true, controlsTutorial: presenter ? false : true, slideNumber: presenter ? null : 'c/t', // The "normal" size of the presentation, aspect ratio will be preserved // when the presentation is scaled to fit different resolutions. Can be // specified using percentage units. width: "100%", height: 700, // Factor of the display size that should remain empty around the content margin: 0.1, multiplex: { url: 'https://p.xobs.io/', id: '17b0ef733e56a4aa', secret: Reveal.getQueryHash().s || null }, // Bounds for smallest/largest possible scale to apply to content minScale: 0.02, maxScale: 5.5, transition: 'slide', // none/fade/slide/convex/concave/zoom // More info https://github.com/hakimel/reveal.js#dependencies dependencies: [ { src: 'lib/js/classList.js', condition: function () { return !document.body.classList; } }, { src: 'plugin/markdown/marked.js', condition: function () { return !!document.querySelector('[data-markdown]'); } }, { src: 'plugin/markdown/markdown.js', condition: function () { return !!document.querySelector('[data-markdown]'); } }, { src: 'plugin/highlight/highlight.js', async: true, callback: function () { hljs.initHighlightingOnLoad(); } }, { src: 'plugin/search/search.js', async: true }, { src: 'plugin/zoom-js/zoom.js', async: true }, { src: 'plugin/notes/notes.js', async: true }, { src: 'lib/js/socket.io.js', async: true }, { src: presenter ? 'plugin/multiplex/master.js' : 'plugin/multiplex/client.js', async: true }, ] }); </script> </body> </html>