That Open, Upgradeable ARM Dev Board Is Trying To Make A Comeback

[starshipeleven]

Seems to me, the FSF has no choice but to turn a 'blind eye' (if it comes f.i. to such mostly open-source-projects), because otherwise they would make no progress at all. This is sad, in one way, but in another way it is just logical, because the FSF is not really in a strong position to force such firms like Allwinner to do/change anything...

No, Allwinner did not violate GPL in that hardware, but in software shipped with their SDK. Since none in his right mind will use their shitty SDK in an opensource project, there is no GPL violation going on there, so the FSF can give it the sticker.

Note that it is not a sticker for allwinner, but a sticker for THIS PARTICULAR DEVICE/BOARD/WHATEVER.

[kaprikawn]

hiya kaprikawn, thanks for asking about this - the question is very similar to that made on the pyra-handhelds forum, here's a link to the question and my reply: https://pyra-handheld.com/boards/thr...3#post-1384732

Thanks for taking the time to reply, your arguments are certainly food for thought. At the very least you've convinced me that it's a little more complex than 'Allwinner has violated GPL and therefore is evil'.

I applaud your efforts. I'm not entirely sure I have a use for your product, but I shall think on it and I'll certainly consider backing.

[lkcl]

lol, laptop for $500 without graphics. why not add $5 and use chip with freedreno support?

dude. i'm almost shocked you've seen the *hardware* level security vulnerabilities that were discovered in qualcomm processors, right? https://slashdot.org/story/16/05/05/...roid-user-data

but you know there's more to it than that (otherwise i would have put in a qualcomm processor already into an EOMA68 Computer Card). to develop a hardware board you need the following things:

* NDA (this can sometimes be optional)
* Truck-load of cash paid up-front for illegal access to GPL-licensed source code (this can sometimes be optional)
* A Reference Design (often these can be found on the internet) with full schematics and PCB CAD files
* A copyright-legal "Board Support Package" of some description (often includes a full OS)
* An Evaluation Kit of some description (often sold at extortionate pricing, but they're getting better these days)
* a Technical Reference Manual
* a full Datasheet
* access to the Power Management ICs (Intel forgets about this)
* access to samples, low-volume quantities, medium-volume quantities, high-volume quantities and to receive pricing on ALL of those as well as lead times and availability
* a guarantee from the SoC manufacturer that they will EVEN SUPPLY YOU (or a way to get hold of the SoCs without their consent, such as off of taobao.com or aliexpress)
* access to tech support (this can be optional)

... by now you should be having that familiar feeling of either glazed eyes, bug-eyes, sinking "pit of the stomach" butterflies or just plain outright shock at the sheer overwhelming number of things needed to have to sort out BEFORE YOU CAN EVEN START to evaluate whether a particular SoC is suitable.

once you've made that evaluation, you can start to do the design work. if you have a Reference Design this will take you about 4-6 weeks, otherwise if you haven't it could take several months. depending on the complexity of the IC (how many pins and how small) you have to shell out either $USD1700 for 5 samples if it's a 4-6 layer PCB or if it's 10-14 layers that could be well in excess of $USD 5,000.

make one mistake in the board layout and you have to do that ALL OVER AGAIN. this is hardware, not software. it cost us ten grand to do the first revision and the guy f****d it up. they charged us 5 grand for the next one because they were so embarrassed.

anyway, let's evaluate one of the qualcomm snapdragon processors.

* NDA: they won't even answer my enquiries, and i don't agree with signing NDAs for libre projects anyway. it sends the wrong message to endorse such behaviour. as a fabless semiconductor company you sell hardware. you don't sell INFORMATION about hardware.
* Truck-load of cash: well, they're not getting that from me, it would be ecologically irresponsible, and i have better things to do
* A Reference Design: not available. so we might be able to reverse-engineer things but it's put the project into "Months to develop" category
* A copyright-legal "Board Support Package": not readily available. Linaro have something remotely approximating a BSP in the form of the "Dragonboard" but only for one of the qualcomm SoCs
* An Evaluation Kit: unofficially available from 96boards - the dragonboard. cheap, too, which is better than most
* A Technical Reference: NOT AVAILABLE. oh dear.
* A full Datasheet: NOT AVAILABLE. oops.
* access to the PMICS: NOT AVAILABLE. ah. so how are we supposed to know how to power this SoC up, then?
* access to samples and volume supplies: NOT AVAILABLE. oops.
* a guarantee that they will supply you: NOT AVAILABLE. oops.
* access to tech support: NON-EXISTENT.

oh and don't forget that this is still a processor that has *hardware* level security vulnerabilities....

... am i painting a big enough picture in bright enough colours a mile high, with the qualcomm head office a tiiiny dot in the front and centre of the picture?

to complain about the fact that i'm using an "old" Allwinner A20 when this... this... minefield of technical and logistical issues has to be walked... meh, come on!!

the funny thing is, that qualcomm processor could actually be considered - there are people out there on aliexpress and taobao selling it, alongside "solder mask" kits for enterprising smartphone repair shops to be able to replace the processor. you *can* get hold of the processor, you *can* get the linux kernel source code, but it's just such a nightmare logistically and a huge risk... i think i'll wait until qualcomm get PROPERLY with the programme and provide PROPER Reference Designs without requiring an NDA.

hope you don't mind...

[lkcl]

Thanks for taking the time to reply, your arguments are certainly food for thought. At the very least you've convinced me that it's a little more complex than 'Allwinner has violated GPL and therefore is evil'.

I applaud your efforts. I'm not entirely sure I have a use for your product, but I shall think on it and I'll certainly consider backing.

thanks kaprikawn - really appreciated. we've got a stack of other product designs in the pipeline (i've been asked at least twice about the "rack mounted co-located ultra-low-power hosting" idea already) - tablets, digital SLR camera (my favourite), hand-held games console (being designed by Manuel) - that one's here http://rhombus-tech.net/community_ideas/games_console/

so if you were to support the project now, you'd have an increased chance that someone would design something in the future that you really wanted. it's an ecosystem based around a standard, not a one-time product in other words.

thank you!

[lkcl]

Not Alwinner, but Mali the issue. With all half-decent chips from Freescale that have a Vivante GPU and other stuff that has a half-working opensource driver that only needs more help to go full-blown support, they choose a crap chip with a Mali.

aiyaaa, freescale's processors - you're talking about the iMX6 right? - use far too much power and are far too expensive. $36 for a quad-core, it draws FOUR AMPS at startup due to inrush current, and it consumes 4.5 watts if pushed to its limit. that's far too much.

this is a $7 processor, it's 2.5 watts flat-out, and it *still* can do 1080p60 video playback. that's amazing!

Still, their laptop is the best shot at a good chassis for a decent DYI ARM laptop.

DIY *libre-licensed* laptop... https://www.youmagine.com/designs/li...-laptop-design

I'd love to be able to design a different modular board for that.

sure - knock yourself out. get in touch via the mailing list, the designs are freely available, i'll point you in the right direction. you'll have to agree to use the GPLv3+ license for the PCB files because that's what they're under, otherwise you can always start from scratch but you'll be doing only 2-layer boards so it's not rocket science.

keep in touch, ok?

[lkcl]

On the FSF RYF edition, it looks like they are physically disabling the Mali support.

physically disabling no, but it turns out that because MALI is memory-mapped and the linux kernel divides memory into physical ram and virtual ram (where userspace only has access to virtual ram), you LITERALLY cannot even SEE the proprietary GPU from userspace - end of story.

the average end-user (and the FSF's criteria are designed for the average end-user **NOT** for the technically-literate people who can make discerning well-informed decisions) therefore won't even know it's there. so, we stand a good chance of successfully getting an "exemption" because of this.

if on the other hand this was an intel processor, where the average end-user could do the GUI-equivalent of "lspci" and "oh look, there's this funny-named proprietary GPU, i wonder how i enable that?" and off they go installing proprietary software and get themselves into all sorts of trouble as a result, that *would* be something we just wouldn't bother applying for because we know damn well we wouldn't get RYF Certification for it.

... make sense?

[lkcl]

The problem, for both this and the Pyra, is getting SoCs in the required quantity of ~4000 when the manufacturers want to sell in hundreds of thousands.

So far the only manufacturers willing to sell in tiny quantities have been TI and Allwinner, which is why most small ARM-board projects have either an OMAP or an A10/A20.

for the allwinner SoCs it's more that the entire industry in guangdong is a chaotic mess, such that the processors regularly "fall off the back of a lorry, mate" (translated to chinese of course). in other words if you want QTY 1, QTY 10, QTY 100, QTY 1000 or even QTY 10,000 there are a truck-load of people - some of them honest some of them not - quite happy to sell you, sting you and generally take your money.

for 100k and above yes you'd go direct to allwinner.

i wasn't aware that TI actually grant access to the OMAP4 and 5 series: have they dealt with that stupid, stupid BXPA export "This is A Weapon Like It's a Big Frickin 100ft ICBM gun barrel, really" thing when the daft buggers are getting the TI processors manufactured in China anyway, duuh!

and, more importantly, are they selling a version of the OMAP4 or OMAP5 *WITHOUT* the insanely-problematic PowerVR graphics? if you're not familiar with the history behind PowerVR's insane problems, you're about to find out, FLHerne... you're lucky to have access to the development team inside imgtec, but i can guarantee you that even *they're* about to find out the historical insane problems associated with PowerVR. it's gotten so complex it's out of control. good luck ok!

[lkcl]

if you are hoping for ARM gaming and similar

.... go get an X-Box

[starshipeleven]

common devboards can't be put in front of the average end-user,

Are you seriously believing in this shit?

I mean, seriously.

Raspberry Pi.

...

...

So, now that the concept has settled I can go on.

I understand that making an enclosed swappable standardized module is a good idea for a long distant future when this catches on and you start shipping this to idiots that can't tell the difference between USB and HDMI, but for a long while you will be dealing with relatively knowledgeable people.

Like those that can change a DIMM module in their laptop or can assemble a PC.

So no, wasting time making your own module is... a waste of time.

i've *evaluated* all the standards i can get hold of... there aren't any that are good enough... so i had to make one!

Which is nice if you have something decent to put in. Issue is, you don't as you aren't a major OEM, and using crappy SoCs isn't going to help you get much traction because Openness isn't going to cover the fact that there is still pretty shitty hardware in that module.

What I would do is making a new standard that I can fit the most pre-made SoM boards into, then make a bunch of adapter boards, that accept a specific SoM and allow to mount it in your laptop chassis/board/whatever.

There are five zillion SoM boards that can be integrated, like the ones used in Hummingboards here https://www.solid-run.com/freescale-.../hummingboard/
the wandboard's http://www.tgdaily.com/hardware-feat...goes-quad-core

And I'm using those as example as the GPU has some kind of open driver for it.

Try do do something like this next time.

[Michael_S]

good question! which i think i will add to the FAQ, so thank you for raising it. the answer is: common devboards can't be put in front of the average end-user, they'll break them by putting it into the pocket of their nylon shirt whilst wearing rubber-soled shoes and walking across the plush fake-fur deep-pile shag carpet (zap, zap...)

second answer: i've *evaluated* all the standards i can get hold of... there aren't any that are good enough... so i had to make one! i've covered all the ones that i can find on elinux.org, http://elinux.org/Embedded_Open_Modular_Architecture and you can read about why the EOMA68 standard has the interfaces it does, in the ecocomputing whitepaper http://rhombus-tech.net/whitepapers/...ing_07sep2015/

but please, if you see a standard or a devboard that *is* suitable, do let me know, okay? because i tell you this, it's a hell of a lot of work to do what i'm doing, and i'll be absolutely blunt and up-front with you, if there was *anybody* else in the world trying to do this they can damn well have my support and get on with it, good luck to them! .. but so far, there isn't... so i'm forced to carry on *sigh*. ehh, c'est la vie.

Thanks for running the project! Trying to make real progress with fully Libre computers has got to be like six knights picking up their lances to do battle with a continent of Smaug-class dragons.

In terms of GPL violations, I would guess that the Free Software Foundation is totally practical - Microsoft, Apple, Oracle, and Sony could collaborate on a device, and release it with top to bottom free software, and the FSF would still certify it without blinking an eye. The statements "This Device Respects Your Freedom" or "This Device Does Not Respect Your Freedom" take into account only the actual device and the software running on it, nothing about the history of the companies involved in its creation.