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

[unixfan2001]

[UPDATE: robclark has been helping review the qualcomm 410, it looks like that would qualify! and is available!]

I'm at work, so I can't google for Futurama style memes. But please do imagine me posting a bunch of those.

Shut up and take my money! Shut up and take my rupees! Shut up and take my bottlecaps!

[SystemCrasher]

TL;DR: Allwinner A20:
PROS:
- Simple & well-understood SoC. No nasty surprises expected.
- Crap-free boot sequence.
- Could be 100% blob-free if you cold live without 3D.
- No foreign code stays online once Linux takes over.
- Real native SATA (ahci)
- Real native gigabit
- Reasonable support by mainline u-boot and kernel.

CONS:
- Just 2 x A7 cores, just 1GHz.
- SATA write speed is slower than expected
- Gigabit will not be saturaed (but beats 100MBps things to the dust anyway)
- Mali GPU is proprietary and crappy.

you don't remember any other platform with enough community to produce videodriver?

Speaking for myself I could live with 2D on most computers, it does not requires Mali usage at all. Just display controller, which got nothing to do with Mali. Sure, 3D is neat, but it is really optional goodie for systems administration, software and hardware development, GPS navigation, browning and so on - the most kinds of activity related to computers use I really care of. Uhm, yeah, 3D gaming is really optional part of my life :P. I'm interested in high-performance computing. But Mali can't do that reasonably. Their GPU architecture is retarded, with no unified shaders, so the best it could do is some crippled GL ES, only used by Android games. I could live without it for sure.

your memory is not very stronghard choice between "it doesn't work at all" and "it works, but has cell modem". let me think...

Sounds like just trolling. But how about some good technical points attached, at very least? If my post looks like personal attack, it is actually wrong and comes down to being a major technical disagreement on systems architecture. Qualcomms are really fucked in this regard and I can't remember anything except Qualcomm using Freedreno. So while I really appreciate Freedreno efforts, it really useless for me since I'm not going to touch backdoored Qualcomm SoCs with 10' pole and your mumbling about 3D sounds like really dumbass consumerism, ignoring a very major core issue about most of Qualcomms.

If you haven't got it, when I've take a look, boot seq of Qualcomm looked like this:
- Huge proprietary ROM kicks in.
- Proprietary ROM reads (signed) early boot stages. You can't replace shady blobs as they are signed.
- Early boot stages are bringing up primary CPU. Primary CPU is a cellular modem.
- Cellular modem firmware (the huge proprietary RTOS blob, signed) starts.
- Firmware brings up system online.
- As some optional step, firmware brings up secondary CPU.
- Secondary CPU starts using locked down boot loader (signed, read by cell modem firmware on primary cpu).
- Locked down boot loader loads Linux/Android. Basically some signed kernel.
- Proprietary firmware stays forever and runs on primary CPU, side by side with Linux. I'm not even sure one could completely shut down cell modem on such architecture at all.

If you'll bother self to inform self what RRLP and Enhanced 911 is, together with getting idea cell modem on primary CPU could access absolutely each and every part of the system, you'll get idea it could get even worse than Intel Management Engine does. Cell modem is inherently backdoored thing, just because these days fascist laws and standards REQUIRE this. It also going to report your precise locasion upon RRLP backdoor request. No auth or confirmations, cell net 0wns you.

i'm even less okay with huge proprietary drivers. so those insisting on mali crap should stop bothering about all-open, RYF and somesuch anyway.

It is pefectly fine as long as one does not needs graphics or okay with 2D only - proprietary part of driver is mostly about shaders for 3D. If one needs 3D: that's an issue!! So feel free to step in and point me a better SoCs. I'll be more than happy to use them, etc. However, boot-stage treachery and cell modem design like qualcomm is IMHO even worse than proprietary drivers, since one could count on backdoors presence and huge firmware capable of accessing all parts of system on "master" cpu is a bitch. And it also inherently possesses network access into cellular network, and going respond to RRLP backdoor requests.

So I do not see what's wrong with RYF qualification as long as one does not brings proprietary 3D driver. Sure, it is fair to warn users 3D is going to be unavailable in this case. Sure it is not a good thing. But other than that Allwinner A20 is really good in terms of preictable, crap-free hardware. It boots from small ROM, there is ROM dump, it is easy to analyze and one could be really sure it just reads your boot loader and GTFOffs forever. Boot loader takes control. Then your kernel, rest of user mode kicks in and so on. This way I could be responsible for system behavior and I really like it this way. Crap-free computing . Even wi-fi could be all open - just use AR9271 @ usb, for example. Yeah, it needs firmware. But firmware is opensource (this design is Atheros, not Qualcomm, Qualcomm designs like ath10k are to be treated with great care due to proprietary/backdoor-like nature of Qualcomm and persistent (ab)use of firmwares to do all kinds of nasty things).

[lkcl]

I'm at work, so I can't google for Futurama style memes. But please do imagine me posting a bunch of those.

Shut up and take my money! Shut up and take my rupees! Shut up and take my bottlecaps!

get on the damn campaign page then ya muppet!

[p.s. update - because we need to get this project off the ground and funded if we want to do more computer cards with better processors.]

[lkcl]

If you haven't got it, when I've take a look, boot seq of Qualcomm looked like this:
- Proprietary ROM kicks in.
- Proprietary ROM reads (signed) early boot stages.
- Early boot stages are bringing up primary CPU. Primary CPU is a cellular modem.
- Cellular modem firmware (the huge proprietary RTOS blob) starts.
- Firmware brings system online.
- As some optional step, this firmware brings up secondary CPU.
- Secondary CPU started, using locked down boot loader.
- Locked down boot loader loads Linux/Android.

hollyyyy crap. that's worse than broadcom. i'd heard that phones were capable of powering themselves on, as well as that they had a "phone home" feature where the NSA or whoever could install whatever arbitrary executables they choose, but it's not until you hear how the SoCs *actually work* does it sink in that those "conspiracy-level" concepts actually could be real.

dang.

so what is it about the 410 that could make it *not* be like this? surely they wouldn't be so dumb as to create an SoC that people could investigate and find out that it contains these kinds of backdoors? i am of course assuming that qualcomm would not be that dumb....

about the AR9271, it was chris from thinkpenguin who was responsible for working with one of the developers who was at atheros at the time, he basically did all the work writing the messages that were sent off to management, then to legal, dealing with their responses etc. etc. working tirelessly until *finally* the full firmware for the AR9271 was released under a libre license. the *actual firmware* running on the chip, *not* just a linux kernel driver "shim" that *uploads* firmware. so i have a TP150 dongle that i'm attempting to get up and running with the EOMA68-A20. anyway thought you might like to know: chris is the sponsor for the laptop project.

[unixfan2001]

I'd like to see some solid proof for SystemCrasher's claims.
After reading the most current remainder of his Mozilla/Rust related rants, I generally take everything he says with a huge table spoon of salt.

[starshipeleven]

*sigh* yeah this is one of those really challenging things that i reluctantly had to deal with, 3-4 years ago. bear in mind that i've been developing and checking and re-evaluating the EOMA68 standard for *FIVE* years.

And yet you still think is a good idea to bottleneck all connectivity of a SoC through what, 68 pins?

Even for shitty simple SoCs like A10/20 you had to make compromises like dumping Sata and ethernet (cool story bro), and for the future you only have a usb 3.1 port and nothing else.

You call that smart and future-proof? I don't.

Non-shit modular hardware design would have been to put each thing on its own connector, like a connector for RGMIII or whatever the fuck is the SoC-ethernet-chip protocol, so that you can either connect there a single eth, a switch chip, or another device that accepts that.

Same for usb, video and whatever else. Even the power supply circuitry should be on a separate board to allow reuse and tailoring to specific use-case

Optimally (don't know how feasible this is) even RAM should have its own connector so you can plop in 2GB, 4GB or whatever, or even swap embedded RAM modules with a DIMM-socket module connected to the same SoC board connector.

This would allow to take modules and swap them to change the resulting device. Of course you would also have to change the firmware as there is nothing that can deal with such low-level changes dynamically, but that's the point.

[SystemCrasher]

hiya SystemCrasher, thanks for joining in here, really appreciated.

Greetings. Uhm, I could be a bit rude/trolly, but at the end of day I'm really up for crap-free computing and care about it :P.

yehhh... the checklist is just enormous and extremely unlikely to be fulfilled, and it's down to the cost-savings of the "fully-integrated" bit that comes with "SoC" - System On A Chip...

World is not perfect, we have to deal with it. And if pal666 blames allwinner, fine. If he could propose better SoC. Just praising Freedreno is cool, but it only exists in qualcomm SoCs AFAIK. Most of Qualcomms are extremely proprietary and troublesome. Not to mention utter lack of cheap SBCs/modules. Freedreno is cool in terms of quality and features, but, erm, "Qualcomm" is a real showstopper. If pal666 thinks he could do RYF cert on Qualcomm, I'm eager to see this show.

i hear ya the nice thing about the A20 is, they didn't bother to put in the usual Treacherous DRM at bootloader time or put in some on-board NAND for a bootloader which is then LOCKED IRREVOCABLY with an e-fuse. amazingly they actually did this for business reasons. recognising that factories in China have ABSOLUTELY ZERO software experience and basically "clone whatever they can get their hands on", they reasoned that to support their customers they would need an "unbrickable" SoC.

Chinese SoCs are interesting by the fact most china companies are putting more efforts into just doing good CPUs and virtually zero efforts in doing treacherous or troublesome or just overengineered crap. That's what I like about them. Relatively easy to understand what's going on and it often happens to be more or less crap-free.

so they designed the A10 (and A20) so that the SD/MMC 0 interface could double-up as a JTAG / UART port,

The stupid thing about AW jtag it only operational after extra init. JTAG often used to inject initial instructions and do whole sequencing, so it could bring entirely dead hardware to life. Not a case for AW though, but since one could just change card or use usb ROM bootstrap, it isn't a big deal.

and could also boot unquestioningly from that very same SD/MMC right from power-on. it loads 16k of "stuff" from SD/MMC directly into memory if there's a "signature" on it (a simple 4-letter text phrase) and off we go.

I could craft bootable SD card for allwinner by bare hands (just Linux and dd), as well as assemble Debian rootfs and I really like it this way. Crap-free systems running Debian are good

*sigh* yeah this is one of those really challenging things that i reluctantly had to deal with, 3-4 years ago. bear in mind that i've been developing and checking and re-evaluating the EOMA68 standard for *FIVE* years. i answer in-depth about this on http://rhombus-tech.net/crowdsupply/#questions, "I'm missing ethernet and SATA", but basically i had to replace SATA and Ethernet with an "up to USB 3.1" option. the reason is: without doing that i would LITERALLY have had only ONE processor that could have gone into the standard in a cost-effective way.

Speaking for myself, there're quite few SoCs I would really appreciate, "replacing" module relpaces cpu, mem subsysem and peripherals. What is left? Bunch of connectors? Hardly most valuable part. Idea of replaceable parts is good on its own. As long as it does not hurts things badly. So far it seems I like Olimex approaches, i.e. if they are up for laptop, they are trying to use laptop enclosure and design set of OpenHW boards which would fit it. Sure, they aren't EOMA68 compatible. But EOMA68 was never meant for laptops. It only reasonable for some industrial applications. Even in deeply emdedded system I may sometimes want SATA or gigabit link. USB 3.x is unpopular in cheap SoCs, hi-end SoCs implementing it are bring plenty of other issues, ranging from poor mainline support to overengineering and treachery. Then price really matters and somehow just SBCs or modules are easier to connect to particular application, cheaper and far more popular/available. It is hard to disregard these facts if I'm trying to get things done. Real world is tricky and there're no perfect solutions. Sometimes one have to make tough decisions.

i'm with you here! and i've been looking constantly for five years! allwinner breaks their SoCs down into two types: those with a maximum of 1280x800 and MALI, and those with PowerVR and much higher resolutions. WHAT GIVES??? come oooon. iMX6 has vivante but has a proprietary VPU. TI uses PowerVR (argh)...

MALI does not even uses unified shaders (except maybe newest parts rarely seen in cheap SoCs). I wouldn't count on this thing being able to be a decent 3D graphics workstation in first place. iMX has got huge on-chip ROM and bizarre pricing. Uhm, price still matters in many applications. On bright side they've at least documented their "secure boot" stuff which is virtually ROM OS. Its wildly overengineered so crafting boot images for iMX isn't easy. But at least they're honest in their datasheets. Unlike some other vendors. Uhm, yeah, Ti, I'm looking on you! Ti shown they're not going to let users into ARM Secure World. They have "secure" and "non-secure" parts. They have ROM entering secure world, this ROM is hidden but it is most privileged part of system. I'm not a big fan of large chunk of hidden ROM code enjoying by ultimate privileges without even option to learn what it does and active vendor's refusal to let users into Secure World. This looks suspicious at very least. Not to mention OMAPs are grossly overengineered and expesive. They could pedal their long term availabilty as much as they want it to, but if allwinner going to run batch of A20s for any reasonable order, it seems Ti has got some truly unexpected competitors on their tail XD.

so yeah ... i think it's time to get an SoC made.

If someone wants to spin off crap-free SoCs, especially opensource, they are really welcome.

[SystemCrasher]

I'd like to see some solid proof for SystemCrasher's claims.
After reading the most current remainder of his Mozilla/Rust related rants, I generally take everything he says with a huge table spoon of salt.

Yeah, I could be oversuspicious about entities who proven self badly in the past. Mozilla vs digital sigs included. So you're right, it is really wise to check my words twice. However, it shouldnt be too hard to check for Qualcomm designs. One could look on boot images and some PDFs & reversing researches around. I'm not creative enough to imagine such a perverted boot sequence. I've just investigated which SoCs are "interesting for me" to chew on. Due to insane amounts of data I could mess or forget something. I'm a mere human, I could be overwhelmed by data. I could take shortcuts like if some vendor does something evil, I would assume further creations of this vendor are also going to be evil, which may or may not be true, but I'm better to be on safe side. Trust is thing which is easy to lose and hard to regain. Things could also change over time. I.e. IIRC there could be some application processor only SoC(s) from Qualcomm, they lack cell modem, so their boot seq should be more sane. But, honestly, after stumbling on Qualcomm ICs using tightly locked down boot seq based on cell modem CPU I've been really not in mood to deal with Qualcomm SoCs at all. So if some data are wrong, blames are accepted, I'll take them into account, etc. I'm trying to be fairly good in understanding boot sequences and system designs, after all. I need it myself to do reasonably what I'm doing and keep going like that.

[unixfan2001]

Yeah, I could be oversuspicious about entities who proven self badly in the past. Mozilla vs digital sigs included.

I've explained this to you in the other thread.
As of current, Firefox add-ons run inside the chrome. The chrome has, for the most part, privileged access to your system resources (with very little protection in its way).
It's not unreasonable to assume that someone would write an addon trashing your machine. In fact, accessing libc and other shared objects is painfully simple, so with some creative coding I could easily trash your filesystem or even just reboot your computer from in there.

It's why Mozilla is getting rid of the current addon system and it's also why they require signing for it.

after stumbling on Qualcomm ICs using tightly locked down boot seq based on cell modem CPU I've been really not in mood to deal with Qualcomm SoCs at all.

Which Qualcomm ICs were these exactly? Do you remember?

I'm trying to be fairly good in understanding boot sequences and system designs, after all. I need it myself to do reasonably what I'm doing and keep going like that.

Heh! You sound like the kind of guy I could need for one of my lower level projects.

[SystemCrasher]

hollyyyy crap. that's worse than broadcom. i'd heard that phones were capable of powering themselves on

This one probably isn't worst part of it. Most systems I've seen were capable of doing so. On mobile devices It usually looks like this:
- There is "power manager" IC, it handles most aspects of system power management like voltages conversion ("regulators") and cutting power off for most parts of system.
- When system is "off", power manager stays powered on but shuts down most of its "regulators". Except RTC power domain (low-freq realtime-clock circuits, taking care of real time) and itself (so system could be powered on after going into "off" state).
- RTC always powered on, keeps track of real time and could act as wakeup source of power manager. Most things do not execute code, it just some set of HW registers in RTC.
- When HW still operational, SW pre-computes next expected wakeup time, like e.g. next alarm or reminder user have set up.
- SW writes next supposed wakeup date/time to RTC wakeup registers.
- While HW is operational, SW updates it from time to time, replacing outdated wakeups.
- Power manager could have other wakeup sources like e.g. usb cable detection to do battery charging, etc.
- Once poweroff requested, all but RTC regulators are down.
- RTC still stays operational and counting, next supposed wakeup date retained in RTC registers.
- Once this date happens, RTC alarm acts as power manager wakeup source, so circuits are powered on, device gets operational and boots OS.
- SW could detect it is a time to show alarm/reminder right now.

So user does not misses alarm/reminder even if device has been powered off. This isn't evil on its own, though one could abuse RTC wakeups by causing unexpected system boots. It comes to question who controls SW in system. And relatively easy to spot. Realistically, nearly all microprocessors and microcontrollers got RTC HW and it behaves like this. I guess one could achieve same behavior on Allwinner, it surely got RTC and AXP20x know not to power off RTC LDO either. Not sure it going to work everywhere to full extent. Say, most AW boards lack RTC backup supply but would keep RTC running as long as main power presents.

as well as that they had a "phone home" feature where the NSA or whoever could install whatever arbitrary executables they choose, but it's not until you hear how the SoCs *actually work* does it sink in that those "conspiracy-level" concepts actually could be real.

RRLP has been mandated by standards for a while, OSMOCOM devs were kind enough to bring attention to the pest - http://security.osmocom.org/trac/wiki/RRLP Also, looking into some random cell modem firmwares gives idea, guess what we're looking for in these hostile worlds of evil. Roughly speaking, RRLP declares cell net could ask cell modem about location and it have to do its best to figure it out and report it back. No user consent asked, etc. Cell net always knows rough location, this one is about PRECISE location (i.e. one seen by GPS).

so what is it about the 410 that could make it *not* be like this? surely they wouldn't be so dumb as to create an SoC that people could investigate and find out that it contains these kinds of backdoors? i am of course assuming that qualcomm would not be that dumb....

Most cell modems are backdoored, roughly since some point of 3G. So if one loves fucked-up architectures, taking SoC with integrated cell modem and GPS is a good bet to win RRLP jackpot. If something lacks cell modem it can't behave like this, obviously. IIRC qualcomm had some app processor(s) without cell modem, but since they're not cheap or widespread I had little reasons to investigate further. Most Qualcomm SoCs I've met were cell modem first, everything else second. If there're some better Qualcomm processors, I could imagine some of them aren't evil. But I'll really leave it to someone like pal666 - I'm better off without Qualcomm whatsoever.

about the AR9271, it was chris from thinkpenguin who was responsible for working with one of the developers who was at atheros at the time, he basically did all the work writing the messages that were sent off to management, then to legal, dealing with their responses etc. etc. working tirelessly until *finally* the full firmware for the AR9271 was released under a libre license. the *actual firmware* running on the chip, *not* just a linux kernel driver "shim" that *uploads* firmware.

I know. I've got some understanding what 9271 is. I've even found datasheet on 9271 somewhere. Internally it mostly Ath9k, like AR9287 IIRC. Though since most Ath9k are HW automations and usb isn't their native IF so they had to use some CPU for usb devices.

so i have a TP150 dongle that i'm attempting to get up and running with the EOMA68-A20. anyway thought you might like to know: chris is the sponsor for the laptop project.

Nice. Though when it comes to laptops I'm not really sure EOMA68 is a good idea. Strongest points of A20 are SATA and Gigabit, everything else is quite boring. Actually I prefer to use it when I need decent net and storage. It isn't bad in other things, but these two are well above of any competitors I could think of.