Hacker News Reader

Comment by ur-whale | Hacker News Reader

The question I have, is: given how maker / oss focused the raspi is, how (and by whom) did they manage to get locked-in and forced to use this proprietary piece?

View on HN

Ben Upton worked for Broadcom during the time of the creation of the Pi. Broadcom gave them a sweetheart deal on the processors, they took it.

The Pi was initially created as a cheap way to get kids into computer science, they didn’t foresee the closed off parts of the processor being an issue towards that goal. They just wanted a cheap computer for kids to learn on that wouldn’t be the end of the world if they broke it. I mean who is gonna want to run an Open Source VideoCore?

THEN, us “grown up” geeks came along and was like “OHHHH, a cheap Linux SBC… Yes Please…” and brought out the initial run on day one.

So ever since then they were kinda stuck with Broadcom unless they wanted to redo a ton with another manufacturer.

To Bens and Raspberry Pis credit, they have managed to get Broadcom more open than they were, Initially we didn’t even have a data sheet for the processor.

Edit: Pre-coffee brain even more prone to typos then when I’m caffeinated…

There must be more to it than that, considering all the generational change otherwise.

I disagree, pioneering SBCs on a non-profit budget means this one hitch on binary blobs is a very low priority for them.

Working with the same vendor consistently to iterate on a very similar design each time is ideal.

Well, that right there does count as "more to it"

> considering all the generational change otherwise

Sorry, my brain is not fully in gear yet. Do you mean the changes between the generations of the Pi?

Eben Upton

My bad, Pre-Coffee brain had Ben stuck in it for some reason.

They just wanted a cheap computer for kids to learn on that wouldn’t be the end of the world if they broke it.

I've always said that an old PC (but not too old, because retrocomputing has driven up prices then) is probably the best for that. Can be bought for next to nothing or even free, has extensive compatibility with lots of software, and also decades of detailed documentation.

That isn't something schools can buy several hundred of (with a stable platform) and not worry about electrical testing liability etc. Which is the main intended use case of the Pi.

Ebon's reasoning at the time was that he wanted a standardized computer you could build a curriculum around, that was cheap enough that schools could issue to kids without fear of them breaking it, and which was small enough the kids could take home with them in their backpack. (This was obviously well before Chromebooks and iPads took over the education market.)

Well there is a benefit to using new hardware, No issues with aging caps, no issues with sourcing peripherals (Unless a pandemic comes along messing up everyones supply chains), no issues with compatability.

By only have a single platform to support out the box you are getting rid of having to support multiple hardware configurations which could cause headaches for newcomers on day one. Remember it was basically an attempt to remove the roadblocks of getting people into CS. One of those roadblocks is getting people to "hello world".

IMO its a simailar reason to why Arduino worked so well, sure we could push people to using any other microcontrollers but by having single known board (atleast to start with) everyone is in the same boat and makes it eaiser (and cheaper) to offer support and lowers the barrier of entry imo. Basically is solving the fragmentation issue.

Is it the best way to learn? That depends on how you look at things. IMO it makes it a great stepping stone to get into the field which can then lead on to other things/interests, but you will probally learn more earlier on by skipping the "spoon feeding" stage but that (imo) comes with a steeper learning curve which could drive people away from the subject.

I know I delayed my own learning of the NRF platform to start with simply because at the time the toolchain was a PITA to get started with (esp on an unclean machine that had other compilers installed) so on a number of times I got fed up trying to get to "hello world" I would put it down and come back to it at a later time. However that process of less handholding did teach me more about the toolchain.

>Arduino

I generally agree on the simple, common approach being a great draw for Arduino and its related education. I was going through school around the time when arudino took off. IMO, older vendor toolchains were just painful by comparison. Licensed compilers ($$ license), janky IDEs that were death by 1000 cuts, having to learn different port masks (etc) for initializing different microcontrollers, IO libraries for each microcontroller, proprietary programmers (devices to load compiled software to the microcontroller). IMO, this is where FPGAs are largely still stuck in nowadays.

Though it probably wasn't all that bad. My experiences with the bad side of things largely stems from the PIC lineup. I still have trusted configurations of MPLAB + C Compiler that work vs others I just could never get working. Still have the PIC programmer. Some earlier arm tooling (armv6 era) was quite like this, too. Luckily, it has all opened up quite a bit. Either arudino-level ease of use or even drag and drop. The latter did exist in the armv6 era, since I have a Freescale Kinetis that operates like that, minus the simple IDE & compiler of the arduinos.

Simple IDE also means simple install, operation, and licensing to me. There may be a great paid IDE for the Kinetis, but the moment I have to start juggling more logins, node/floating license files, web-only environments, etc, I just remember it as time wasted on superfluous nonsense.

For me these days I value "Time to hello world" over many other things which is why I would rather use PlatformIO when I'm using a platform & framework it supports even if its lagging slightly behind the latest framework version from the vendor directly.

But looking back, back in the day when we had to walk uphill both ways in the snow to compile and write (get off mah lawn! :-P) I'm grateful I did learn "how the glue was made" instead of just using something ready made. But older grumpier me just wants to get shit done so I'm happy those days are pretty much behind me, but I'm ready to dust them off again if it was really needed.

The Raspberry Pi was basically only possible because of the Broadcom SoC. It is a capable chip, with approximately the right set of functionality for an SBC and at a good price (because it was already manufactured in volume). It was only available for something like the Raspberry Pi because Eben Upton worked for Broadcom and so could get to buy the chips and buy them at volume pricing - normally you'd have to have an established buyer relationship and be able to guarantee to buy far more than the Pi team was able to or expecting.

It's also worth remembering that originally the Pi wasn't maker/OSS focused - the goal was to have a computer cheap enough to be used for computing education in schools. In effect a modern day successor to the BBC Micro.

In the context of the goals and constraints the "minor" binary blob required to make it run was irrelevant. Even more so as basically every other similar SoC has exactly the same issue. The Broadcom parts presumably continue to remain competitive for their capability level and so they keep getting used but now thanks to the success of the Pi there is the will and capability of going full OSS.

That's because the idea for the RPi came from Eben Upton, who worked at Broadcom at the time - and no one else had better chips or (and here it gets crucial) would make them available at the low quantities that were initially expected without serious up-front money to get access to technical documentation and experience.

Creating a computing platform is - no matter the CPU vendor - one hell of an effort, often involving bunches of binary blobs of questionable quality, NDAs, buggy, outdated or plain lacking documentation and lots of money. The more effort you can save yourself (such as by using a product you already have experience with), the better.

RPi1 was also built around a surplus SoC that was originally targeted for a set-top video player platform but failed to get any buyers.

By the entire industry. There simply doesn't exist a chip that does the things a raspberry pi needs, that is free of proprietary software.

RISC-V is probably the closest, but they're going to need to get faster to compete with ARM.

As it's the GPU that seems to be the issue here is there a RISC-V based SoC with an open source GPU driver?

Worth noting too that there is an open source driver for Arm's own Mali GPUs.

https://www.collabora.com/news-and-blog/blog/2021/06/11/open...

The GPU is not the issue. There are open source OpenGL ES drivers available in upstream Mesa right now, that work with upstream kernel DRM driver, and for RPi4 users there's even a Vulkan driver.

It's been pointed out elsewhere but briefly there is a VLIW processor (the "VPU") that is initially in charge of the entire boot sequence before handing off control to the main ARM cores; the bootcode.bin firmware for RPi devices is exactly this code. This includes things like bringing up PLLs and the on-board UART before handing off control to the ARM core where "userspace" code runs.

There are many free RISC-V implementations, and several free GPU drivers for various hardware families, but there is no combination of the two in any meaningful sense right now. If I had to guess I'd say ImgTec is probably one of the ones you could expect to pop up in an SoC somewhere, since I doubt ARM or Qualcomm are going to license their GPUs outside their families... ImgTec recently started contributing some code to Mesa but otherwise have historically been pretty hostile. So the immediate speculation doesn't look great at the moment but who knows what could happen.

The VPU isn’t a VLIW at all, it’s an interesting RISC.

Arm GPUs are also licensable by anybody, they were x86 phones with them back when those were still a thing.

Interesting, thanks for both the corrections. Curious, but do you know if there anyone still using Mali outside the ARM-licensed family at this point? I guess there haven't been many new entries to the mobile market for so long the GPU pairings seem natural, now...

Isn’t it right that there’s barely any hardware at all out there that can run fully foss?

If running without blobs is your priority, various Allwinner or Rockchip SoCs would be much better choice. (they run with upstream kernel, and aren't GPU with ARM core added as an afterthought)

beaglebone

Not true at all. Pinephone runs on a SoC with no binary blobs user has to install.

And the SoC can do all the stuff my Rpi 2b does, and more (has a proper audio codec, for one, real gigabit ethernet, working suspend to ram, etc.), does suspend to ram, accelerated video decoding, and has much more open documentation... Rpi 2b's soc is pretty terrible usability wise comapred to A64.

It is actually the other way around since a lot of other boards use no proprietary blobs, either out of the box or can be easily adapted to do that. The RPi Foundation are without doubt good at making hardware, but they're even better at letting users believe they're the only player in that field.

Because frankly the parts available with fully free internals are...not very good. How many people care about the raspi having decent performance vs how many people care about the binary blob?

As I understand it, it was a SoC that was already in production by Broadcom for set top boxes so using an "off the shelf" SoC would reduce the time/cost required to bring the original Pi to market. I imagine there would be less risk to the manufacturer in this case since, if the Pi proved a failure and didn't move the expected units, the SoC could be repurposed for STBs.

My understanding was the raspberry pi was supposed to be the modern zx80: cheap above all else to make it easy for kids to get a computer.

And it pretty much is, especially the older versions of the Pi. It just has a large number of uses beyond education.

Reply on news.ycombinator.com