DrYak | # 191 | 2017-08-01, 13:28 | Report

Beside the GPU, there are other driver you need to take of :

- Cell network modem :

There used to be a time when this was a discrete chips that only communicated over a standard link mimicking a serial channel to a modem speaking standard "AT" command set.

(That's how the OpenMOKO was designed back then. That's how the Pyra is still designed today).

But for energy/spacing efficiencies (and also to make the NSA happy one would add) things nowadays are different.
The cell network modem is usually built-in directly in the SoC.

Sometimes, the cell core even serves as THE SOC'S NORTH BRIDGE, and is in charge of RAM, ROM, etc.

Making things even more complicated is that, for radio-licensing reasons, the firmware can't legally be opensourced, and some parts are under the responsibility of the cell service provider.


- Wifi and Bluetooth are other cores that need drivers.
(But I am under the impression that most of the cores are also popular in other contexts and lots of them DO HAVE mainstream linux support).


- Cameras are another problem.

On chromebooks/netbooks, just like on desktops, they might be standard USB parts, most of which nowadays are supported by Linux, specially if they conform to the UVC class.

On phones, well, it's a different matter.

- Then there are all the sensors (GPS, accelerometers, touch)

A netbook could use more standard parts (a GPS that simply talks over a serial channel) or circumvent (if the touchpad isn't supported yet, just plug a mouse in).

On a phone less so.

DrYak | # 192 | 2017-08-01, 13:31 | Report

So yes, one could run Arch on a phone, even today.

But you'll be basically stuck with only having a serial console over the SoC's UART, probably some USB networking (or at least a serial-over-USB console), and maybe if you're lucky some Wifi networking.

Everythink else (screen, even leds, camera, sensors) staying desperately un-powered.


(Basically like when playing with some quick port of a distro on the OpenMoko back then).

DrYak | # 193 | 2017-08-01, 13:45 | Report

If you want a functionnal Linux on your device, it's not going to be cheap and easy.


Solution 1:
would be to adapt libhybris to get it working with the distro so the phone could use the Android drivers.

Ubuntu Touch was the only desktop-style distro doing it on a large official scale. But they've quit.

Gentoo and Arch being hacker friendly should be the easiest, but one would need to pay devs to do it.

Other distro might have devs and might get paid to do it (I know that opensuse has shown interest in ARM netbooks/chromebooks in the past).

The good thing : you can release such a device now with only Android and SFOS support, and add the Linux distro later when it's working.



Solution 2:
- redesign the device to use only parts with at least official linux support.
(That's how most netbooks/chromebooks did it.
Pyra is partially going this way, regarding the GPU)
That's going to require some redesign (probably custom PCB) and select parts which aren't optimal (different power drain needs between chromebooks and smartphone)

And by the time the things is finally done, released and start shipping, users will complain that its specs look outdated compared to the latest iPhone Galaxy 9 Plus whatever.

The only reason that Pyra is managing to ship at a not to glacial rate is that they could leverage a lot of past work (e.g.: the main board is actually an evolution that Golden Delicious did for the Openmoko upgrade "GTA04") combined with Texas Instrument OMAP SoC having pin-compatible variant (they could basically replace the SoC with the latest available while keeping more or less the board).



Solution3:
- have paid developers do the reverse engineering for the GPU
- use as much standard component for the rest
so any modern kernel could work on it.
- design a completely custom solution around the above.

This is tremendously complex work for a phone.

(The only one I know who managed to pull it was the OpenMoko.
And it still took ages before coming out and looked outdated.
And they used very carefully crafted platform (both GPS and Cell-modem are completely self-contained chips that simply talk to the main system over a serial COM port.
The GPU has a weird architecture that can completely be short-cut around a 2D rendering done while the 3D was only closed soruce.
The Wifi and bluetooth are on purpose standard cores that have similar desktop drivers.
etc.)

DrYak | # 194 | 2017-08-01, 13:50 | Report

So for now, chen is better of :

- keeping whatever chip he think would be the best easiest for him now.
(From what I've understood : basically a chip with decent spec, availble in low volume at a reasonnable price, with a ready-to-use PCB design)

- optionnally provide Android for non-power users.

- pay Jolla to provide SFOS with libhybris.


set a stretch goal to pay developpers at - e.g. OpenSUSE - to make a version of tumbleweed that can work over libhybris.
(or see if there's a community that picked up the ball after Ubuntu Touch and can be paid to adapt it to the phone).



Maybe at a laster point design a future phone with more linux friendly parts (hey, maybe by then AMD will release a low-power version of their mythical upcoming ARM CPU. Or maybe Nvidia would have released enough docs for Nouveau driver to be good on Tegra).
With USB and I2C pins :-D

TheKit | # 195 | 2017-08-01, 15:19 | Report

Having libhybris with Arch is mostly a packaging matter, but then you need software which can handle it. For example, only Plasma Desktop/Mobile can be currently used as desktop environment (since they implemented hwcomposer support in KWin-wayland as part of Plasma Mobile project).

I wrote about graphics support regarding Maemo porting before, but even after everyting written in this thread it's still complicated.

Qualcomm has Freedreno open-source driver, which is good, but it's only actively used with DragonBoards and developed against mainline kernel, which you can't run on phone easily due to drivers for other components missing.

MediaTek has Chromebooks, which run on kernel, close to mainline (compared to Android), but the SoCs in them are usually used for tablets, not phones, and they still employ closed-source PowerVR blobs for graphics.

Then we also have less known AllWinner and Rockchip SoCs, which have pretty good mainline support for some of them (to the extent AllWinner tablet was tried as reference device for Maemo porting), but I didn't see any phones based on them, which likely means they don't have modem solutions. Mali blobs are used for graphics (Lima is not in usable status and not actively developed).

---

Last edited by TheKit; 2017-08-01 at 15:25.

chenliangchen | # 196 | 2017-08-01, 16:16 | Report

Thank @DrYAk @The Kit @Venemo @locusf and @zeta for all the information. I need to study them...

BTW I have updated the title and the 1st post, to make the discussion matching the topic.

Also we have kicked off some structural design, and work on changing PCB. (@DrYak To answer your previous question, the PCB will be changed slightly to fit the keyboard connection and a hall effect sensor for the slider.)

Structural design is also not easy on this one... I want to maintain the looking that we loved, and also make it metal.

Will share more pictures through out design progress. Keypad will be re-designed as well. From feedback of Moto Keyboard Mod, the majority seem to prefer offset QAZ keys like PC rather than them on top of each other, the Nokia style...

Attached Images

 

---

Last edited by chenliangchen; 2017-08-01 at 16:27.

Kabouik | # 197 | 2017-08-01, 16:54 | Report

Glad to read explicitly in the first post that Sailfish is the first priority, not an extra. Not sure if you meant official Sailifsh with AD or a port, but it's already sending good waves. Glad to see this mockup as well! Looks very familiar, it seems that the reference to the Lauta was not chosen ramdomly.

About the metal casing, three naive questions here:
- Does it allow thinner casing than, say, polycarbonate (or whatever was the synthetic material used on the N9, and maybe the Lauta if I remember the pictures correctly), and therefore smaller overall size? The reason I'm asking how it compares to the N9 material is I remember it looked great and felt even better in the hands, it also aged relatively well in terms of signs of wear.
- Does it provide better shock-absorption and therefore better fall-resistance of the screen? Phones do fall, that's a law of physics. But the more they can survive a fall, the more they can fall again over their life. And more falls is better. Of the three questions here, this is actually the most important to me; I would not want to witness such a unique but limited phone break. All my phones so far have been very resistant, never broke a screen, but they were all plastic only (and in general, removable back cover seemed to significantly increase shock absorption).
- Is metal considered by Livermorium an important marketing argument, in addition to putative benefits above? This is not a trap question, marketing is important and I know nothing about it. If something is objectively inferior but makes a device sell better, then maybe it's better to the overall project (but there are limits to the idea, like Android instead of another OS ). If so, is it considered the only material that would make the phone look good, or are there other candidates (like N9 material for instance)?

A little bit surprised, but to be honest I wouldn't even know what would be my preference if I was asked. If I imagine myself thumb-typing on keys that are almost completely concealed by my thumbs, I would assume I need a pretty good tactile feedback and that would be what matters the most.

If keys are dome-shaped like on the N900, and therefore easy to distinguish from top row to bottom row, then I think aligned rows are less confusing; very few tactile information needed to discriminate them and the spatial relationships between keys are simpler, it's just a grid.

Another thing to keep in mind: thumb-typing implies typing not only with the middle of the thumb, but also with the sides just by twisting the thumb left and right without actually moving its position on the keypad. This was perfectly achieved with the N900 and dome-shaped keys, the thumb could just navigate in columns and rows through that grid, sliding in straight lines between the keys. This was a very easy way to orientate it and always know approximately where it was, and one could type either Q, W, A, or S without changing position. I'm afraid a keypad with offset would prevent sliding the thumb and typing four keys with minimal move like this. Would it be harder to type? I cannot tell, I don't know. It probably partly depends on the overall size of the keyboard, keys and gaps as well.

I will trust whatever you guys think is best, you have the experience, but please don't try to mimic full size keyboards just for the sake of mimicking them without making sure it wouldn't hurt thumb-typing experience.

---

Last edited by Kabouik; 2017-08-01 at 17:13.

Metsämies | # 198 | 2017-08-01, 17:15 | Report

I had both N9 and E7 and N9 had better data connections/network quality in poor signal areas. So polycarbonate is not bad option!

chenliangchen | # 199 | 2017-08-01, 17:30 | Report

It uses different type of antenna than N950 and E7.

gerbick | # 200 | 2017-08-01, 18:09 | Report

Curious question... but why would you? It would probably cease being a phone and now is a device with a lot of access points (wi-fi, GSM radios, bluetooth) that more than likely will not work.

Or am I missing something - or a lot of something?