Hacker News new | ask | show | jobs
Rust on RTL8710 running FreeRTOS (polyfractal.com)
245 points by aurhum 3448 days ago
7 comments

It's going to be really great as Rust spreads further and further into the embedded world. I know enough C to get by, but I'm always second guessing myself when things start to get complicated, and Rust has really made lowish-level programming a lot easier for me. The safety features in addition to exposing a C ABI make it a great language for build once, run everywhere (everywhere that LLVM supports, anyway). Exciting times.
Unfortunately, there's little chance for Rust to make inroads into the embedded market, tho I wish it were otherwise.

C++ has had many benefits over C for embedded s/w for many years and has similarly failed. Note that contrary to popular belief, C++ does not result in larger, more inefficient s/w than C (for the correct C++ subset).

The momentum behind the C ecosystem is so overwhelming that Rust simply will not get a foothold anytime soon.

That's why I quite like the approach of using existing C tooling and SDK, then linking in "user-land" Rust. I wanted to run stuff on the RTL8710, and it seemed crazy to start rewriting all the HAL from scratch.

But it only took an afternoon to bolt Rust into the C build because Rust's FFI story is really nice (imo). So I get the best of both worlds: existing C HAL and SDK with the pleasantness and safety of Rust.*

* With the assumption/hope that the HAL isn't buggy, since Rust will rely on that to not explode :)

Brilliant! Thanks for the proof of concept. I've been debating this method, but it looks like you got it setup relatively quickly. Almost as easy as transitioning to another C compiler/tool chain.
> The momentum behind the C ecosystem is so overwhelming that Rust simply will not get a foothold anytime soon.

You could also say this for C on Unix in 1990. Or x86 assembly for PCs in 1985. In 1990, the idea that, 25 years later, we would be deploying network services written in JavaScript was unthinkable.

Change takes time, and C won't die, ever—but history shows that change eventually does happen.

> You could also say this for C on Unix in 1990.

You can still say this on 2017.

No UNIX kernel will ever be written in anything other than C, the way they are married to each other and the way UNIX culture works.

To get rid of C we need to get rid of UNIX, even a POSIX like OS coded in say Ada, needs to expose C like semantics for POSIX compatibility.

Which means we still have a long way ahead in what concerns improving security.

Yeah, I was referring to the userland, not the kernel.
My company uses a C++ subset (no STL, no malloc) on all of our embedded projects, which are mostly on ARM Cortex-M4.

I'd love to learn and start using Rust in our work, but we are client-project-driven, so it will be hard to justify the investment and productivity drop. I think it would make it easier to attract top talent, though things are weird in the embedded world.

I had a co-worker a long time ago described this problem as being a bunch of lumberjacks chopping down trees, but who were never allowed to stop and sharpen their axe's (let alone procure the chainsaws and learn how to use those...).
I'm a C++ enthusiast interested in embedded programming but I have very little experience. What is the rationale for your company's avoidance of the STL, is it primarily binary size?
Likely because it allocates and uses exceptions pervasively.
We use embedded C++ though it is mostly C with classes and sugar after throwing out STL and exceptions. I'm eagerly awaiting more embedded friendly Nim improvements since it can be plumbed into existing embedded C tool chains.
Happy to see that you mentioned Nim. What improvements are you looking for in regards to embedded development?
The ability to do useful things without the GC would be nice.
What makes the ecosystem so unfriendly to languages that link to C easily, either way?
The ecosystem... from people to compilers to libraries to tutorials... is all C and its style of doing things. Gotta be as close to it as possible to have a chance. It's why the successful languages in enterprise, system space used C-like syntax. Still took a huge company basically dictating it.
The culture, the outdated compilers (expect C89 and C++98) and software designed by hardware engineers.
What about ARM's mbed bet(and ARM know their stuff) , written in c++ ? no chance for it sucseeding in the IOT , yet ARM claim a large ecosystem ?
mbed is not a large ecosystem, its large in the 'IoT' world, but no one doing real embedded work (more than say half a million units) would use mbed.
>> real embedded work (more than say half a million units)

Altough most of the units shipped will be above this, won't a decent share of projects(in general ,not just the IOT) be below that threshold ?

Is rust actually used in the embedded world, anywhere? All sorts of languages have been made to run on micro controllers, from Pascal to Scheme.
Got talking to a dude on the phone (random call about something totally non-technical). Turned out he was an embedded systems engineer in the Nuclear industry. Uses Forth for everything. He strongly encouraged me to use it for any embedded hobby project.
JeeLabs has some stuff on Forth in the uC world: http://jeelabs.org/2016/02/dive-into-forth/
I love that jeelabs blog.. Also, some languages running on esp8266:

https://github.com/yesco/esp-lisp

https://github.com/zeroflag/punyforth

https://micropython.org/

(and of course lua which i didn't try because its official... https://github.com/nodemcu/nodemcu-firmware)

damn that looks fun. as if I didn't already have enough things I want to learn...
Charles Moore?
Not many languages can be made to run efficiently on a 83 MHz core with few kbytes of RAM.

Pascal, maybe, but it's about the same thing as C.

Scheme would be nice, but it's both more resource-hungry and the static checking is sort of bolted on.

With Rust's security features I hope we'll have fewer crashing or crackable embedded devices.

Pascal, these guys have options for almost all relevant processors.

http://www.mikroe.com/mikropascal/

Scheme, do people actually know how powerful RTL8710 is in regards to the computers used to create Scheme?

RTL8710 has a ARM Cortex M3 @ 166 MHz with 48KB available to user and 1MB flash.

Scheme was developed on the ITS OS, running on PDP-10, which could have up to 256 kilowords (not KB), and ran at about 30 MHz.

I am with you on Rust.

The majority of these tiny devices have more resources than 70's mainframes, that used safer systems programming languages.

Tl;dr: this works because the RTL8710 is an ARM core, which is supported by LLVM and thus Rust anyway. Its a nice how-to for working with the chip though.
Just because something uses ARM doesn't mean it'll be easy to target, there's a hell of a lot more to getting something running on an embedded system than just what basic architecture it supports.
Soldering tip: tack a pin down on opposing corners of the chip before attacking the rest.
Ah, yeah, I can see how that would have helped tremendously. Thanks for the tip! :)

I cringed when I saw this at the top of HN because I knew the shame that is my soldering skill would be seen by a few hundred people :)

If you have analytics on your blog you'll probably find a bit more than a few hundred. Front page #1 stories often accumulate thousands of views an hour.
Been there done that myself. My early works are masterpieces of kapton-tape strain relief because the joints were so bad.
Painfully obvious – and I never thought of it myself! Thank you
Delighted to see this tidy summary! I've been wading through four or five documentation sources over the past few days getting everything hooked up. I'm working on a project that uses SPI and would love to use Rust... I may pick your brains further.

Strongly recommend you post a link to this at the PADI Stamp forums, I'm sure others will appreciate it. http://forum.pine64.org/forumdisplay.php?fid=57

Will do, thanks for the link! Didn't realize the Pine64 folks had a forum setup.

Fair warning, my hardware skills are shoddy at best, but I may be able to help with the Rust side :)

This is cool, though not having the standard library would be painful. Hopefully, there is some middle ground than jettisoning the entire std crate.
That's what libcore is; you still get a lot of the useful bits.

Most of the rest of libstd requires a known OS abstraction to exist.

Most of the algorithmy stuff (e.g. regex) has been moved out of the stdlib so you can pull it in separately if the crate supports no_std.

Not perfect, and a lot more crates could support no_std, but it's sometimes good enough.

What is an example use for this chip on its own? Can it be made to run on low enough power to run on a battery?
Most people use these little things for sensor/data loggers (weather stations, power monitoring, indoor temperature, etc) or little widgets that need connectivity and can be tethered to a power outlet (small display showing the weather outside, etc)

The specs say it has 0.9 mA light sleep, and a 10 uA deep sleep. Assuming the specs are correct (I haven't test it yet... not sure anyone else has either) you could definitely run it off a battery if you aren't waking it up too often.

The real problem with the RTL8710 (and ESP8266, and any other chip that uses WiFi) is that WiFi is very energy intensive. It takes a lot of juice to blast out a 2.4Ghz signal.

I don't know what the numbers are for the RTL8710, but the ESP8266 uses 50-170mA when receiving/transmitting data, which will drain a battery very quickly. (Source: http://bbs.espressif.com/viewtopic.php?t=133)

Pretty much all WiFi transceivers are indeed power-intensive, but it is actually receiving that often takes more juice.

ESP8266 consumes ~70mA with both CPU and WiFi running but idle, and 15mA with CPU idling and WiFi powered down [1]. That's 55mA or (@3.3v) ~180mW for _idle_ (i.e. receiving) WiFi circuitry. Again - that's WiFi only, with exactly zero RF power transmitted.

Now transmit. EU restricts amount of RF power radiated by a WiFi station/AP to 100mW. In US and few other places you can go slightly higher, but then other challenges arise, so for practical purpose most WiFi chipsets top out at 100 mW (a.k.a. 20dBm). Assuming RF power amplifier has efficiency of 32% [2], transmit duty cycle of 50% (that's transmitting half of the time and receiving for another half - pretty intense traffic here), we get:

(100/32%)*50% = just 156mW consumed towards "blasting out" some RF.

Clearly, receiving WiFi takes more. And there is a good reason for that. Most modern designs implement Rx as: Low-noise amplifier with some bandpass filtering -> Zero-IF mixer -> quadrature ADC -> DSP. It's the latter that does FFT, I/Q constellation tracking, demodulation, forward error correction, etc. essentially in software. And that takes a lot of power.

[1] http://bbs.espressif.com/viewtopic.php?t=133

[2] http://ww1.microchip.com/downloads/en/DeviceDoc/75003C.pdf

Ah, neat. Thanks for the analysis! I always assumed it was the transmit that ate the power, simply because you need `x` amount of energy to broadcast the signal out. But I can absolutely see how it's the circuitry associated with getting a signal back into digital that could be costly instead

TIL :)

thank you. I suppose the critical factor is, how long and how often does it have to wake up, and more importantly, how many milliseconds of wifi power is needed to transmit whatever data was gathered.
http://aitendo3.sakura.ne.jp/aitendo_data/product_img/wirele... says 180mAh when actively transmitting, so probably.
Nice, didn't know about this chip.