Reduce and simplify: ubuntu armhf img

Sorry, i said i will return to my normal life... i hoped to ;-)

But i got a wise comment around the fact all this crap with multiple 'dd', spl, uboot, bootstrap the distrib, etc.. is just hell for 99% of the guys wanting to play with the mele under linux.

And it's just true.

99% of mele owners don't give a sh** of the technical masturbation around u-boot, nand and all other things.

As asked, i have made a quick and dirty ubuntu image containing the kernel (sata, wemac tuned), working u-boot, and nand install... all ready to be 'DD' or 'Rawwrite' to a MMC from linux or windows (didn't try rawwrite).

I have post the image on my website and hope someone else will follow this path to simplify the whole install for everyone: I'm ready to host anyone files that will follow the same spirit but i'm sure Jeff could do it.

http://guillaumeplayground.net/mele-a2000-ubuntu-armhf/

No MMC partionning, No multi package download, No compilation and a simple nand install.

I know this image is really stupid, quick and dirty... two hours work.
In fact, i'm not an ubuntu fan - because i think it's simplifying too much debian ;-) and in a way makes the user less skilled... but making things simplier is the way for mele.

Users who want to learn linux internals will do it on their desktop... and mele owners (this community) is just hungry/angry to get a working linux on their mele.

Hope it will help and initiate a new way to get users ready to run with their mele.

Quote
Tony Stark
That's how Dad did it, that's how America does it, and it's worked out pretty well so far.

Regards,
--
Guillaume

Finally, someone had to produce something...

Ubuntu 12.10 Armhf with kernel 3.0.42+.

New version with xorg working:
http://guillaumeplayground.net/mele-a2000-ubuntu-12-10-armhf-base-with-nand-install-v1/

And my addon for the community: PimpMyMele.
Nand install, screen output configuration and led blinking ;-)
http://guillaumeplayground.net/pimp-my-mele/
I will release it separatly later.

Have fun.
Bye.

Thank you so much Guillaume, going to try it in the weekend :)

by the way: is it working with v1.7 too?

yes, v1.7, v1.3 etc... it's the release i used in my kernels with:
- sata clock patch i have work on (the one in amery tree)
- wemac tuned (not the slow one they put in amery tree).
- uboot clock patched.

And working great so far on my cluster (mixed between v1.3 and v1.7):
$ uptime
13:03:26 up 29 days, 19:00, 1 user, load average: 0.19, 0.26, 0.30

Bye.

Guill, have you tested the wlan0 (8192.ko module) to Soft-AP using hostapd?
Can you try to compile the hostapd for 3.0.42 kernel?

(guide: http://jas-hacks.blogspot.it/2012/09/hackberry-wireless-access-point.html)
(last driver RTL8192CU: http://www.realtek.com/downloads/downloadsView.aspx?Langid=1&PNid=48&PFid=48&Level=5&Conn=4&DownTypeID=3&GetDown=false&Downloads=true#2312)

Thank you

and possibly disabling all power save featues:
rtw_power_mgnt=0
rtw_enusbss=0

You try ;-)

wlan0 (8192cu) is working.
eth0 (wemac) is working.

hostapd is simply a user-land software using a bridge and managing WPA.
It should work.

You mention you prefer Debian - do you have any Debian ARMHF images around you'd care to share as well?

~Jeff

Image cannot run via SD on my Mele...

Claudio Wrote:
-------------------------------------------------------
> Image cannot run via SD on my Mele...

Should work ;-)
What kind of Mele do you have ?
How did you inject the gunziped img on the SD ?
Is your SDCard correct ?
Did you try to unplug the power during 5-6 sec and retry ? (sometimes on some mele, it's failing to detect the SD and boot to android).

In the Uboot/Kernel boot, the led is Red/Blue.
In the Userland init, the led is Blue.
And when it's finished: it's making a nice flap between red/blue and stabilize to blue (this one is configurable) ;-)

JeffHoogland Wrote:
-------------------------------------------------------
> You mention you prefer Debian - do you have any
> Debian ARMHF images around you'd care to share as
> well?
>
> ~Jeff

I have an armhf debian testing but it's an heavily modified and puppetized version with my internal apt server.

I could prepare another debian testing headless with the full network stack (ipv6, ipvs, tap/tun, bridge).
Debian testing is nearly always up-to-date with latest versions and pretty stable and reliable.

n.b. if you are a windows guy (like me) theres a great really simple tool to backup and restore sd card (and usb stick) images - i use this a lot to back up and restore various armhf builds (i use my mele 24/7 now as a squeezebox/minidlna headless server but still play with it by swapping sd cards)

- you just have to remember to right click (run as admin) when you start it
- as sd cards os the same GB size have usually different unrounded sizes theres a useful option to "truncate" which allows you restore a bigger image on a smaller SD

usb image tool

http://www.alexpage.de/usb-image-tool/

guillaume Wrote:
-------------------------------------------------------
> Claudio Wrote:
> --------------------------------------------------
> -----
> > Image cannot run via SD on my Mele...
>
> Should work ;-)
> What kind of Mele do you have ?
> How did you inject the gunziped img on the SD ?
> Is your SDCard correct ?
> Did you try to unplug the power during 5-6 sec and
> retry ? (sometimes on some mele, it's failing to
> detect the SD and boot to android).
>
> In the Uboot/Kernel boot, the led is Red/Blue.
> In the Userland init, the led is Blue.
> And when it's finished: it's making a nice flap
> between red/blue and stabilize to blue (this one
> is configurable) ;-)


Is that any way to get the Mele version (1.7, ecc) from os command?

I've build the SD in the right way, and SD is not broken. I think the problem could be the kernel 3.0.42.

Thanks guillaume for making our life easy...

> Is that any way to get the Mele version (1.7, ecc)
> from os command?

There are no tools to do this yet, but it should be possible to build a script that could extract and decompile the system.bin from the nand and look for values that are unique to each mele revision. If script.bin does not have unique values then script could compare the md5 hash of the nand bootloader files against a known database.

The foundations for such a tool are already build into my rescue system. If I can get some volunteers with different a10 devices, I'd be happy to develop this further.

Quote
Claudio
I've build the SD in the right way, and SD is not broken. I think the problem could be the kernel 3.0.42.

In fact, it seems the mali driver make the ubuntu stuck for some mele versions...

If you are on linux on your desktop, you can try this:
$ mount -oloop,offset=15728640 /dev/YOUR_SDCARD /mnt/
edit /mnt/etc/modules and comment all modules except sun4i_gpio (the led).
$ umount /mnt
take back your SD and try to boot it from mele. It will boot.
Once on the mele's ubuntu in ssh, modprobe each one in the same order to see if one is blocking.
It should be the 'mali'...

I will try to find why, i need to take back some other mele from my office before (tomorrow)... but i only have lastest versions left.

Quote

I think the problem could be the kernel 3.0.42.

When I tried the 3.0.42, posted here by me, last night on a tablet I noticed that video output was not working. There are some changes in 3.0.42 framebuffer configs compared to the earlier versions. I have them configured as modules to better cooperate with Mali support. But it appears they do not load in time to work correctly. The LCD, HDMI and other video drivers need to be built in. The config will likely need to revert to using built-in video modules as in 3.0.39. I can't do that until I get back home in about 10 days.

In the meantime I recommend using the 3.0.39 kernel. (is that even posted here?). Maybe somebody else could post a version with working LCD, HDMI, at least until I get back home and can recompile.

Sorry about the oversight on the vid. I no longer have easily accessible devices to test on.

Quote
gnexus
When I tried the 3.0.42, posted here by me, last night on a tablet I noticed that video output was not working. There are some changes in 3.0.42 framebuffer configs compared to the earlier versions.

If there is changes, did you reverse them ?

This version is working on my mele(s) and i will have to get a VersA/B back from my production to try it. That's what i have plan to try to do next week.
Unless someone could try some tuned images?

I still have some doubt about the omni-version clock patch forked from the original clock change (from the sata patch). It could be good to verify it's not linked... i will build one and post it.

Quote
gnexus
In the meantime I recommend using the 3.0.39 kernel. (is that even posted here?). I can't do that until I get back home in about 10 days.

Gnexus, i'm sorry but downgrading was and will never be the way.
This kind of process is not making the science move forward.
It's just blind hope that someone will one day find the problem for you.



Edited 1 time(s). Last edit at 11/03/2012 08:39AM by guillaume.

Hey gnexus, I have tried version 3.0.42 from my own build on a tablet.. Works fine for me... No extra config parameters were set.. You want the script.fex/ kernel config??



Edited 1 time(s). Last edit at 11/05/2012 12:08AM by johnsonc.

The latest version -t0 is reported to work on all mele (VersA/B/C) with mali.
I will need to return to Amery this kernel problem.

Just wanted to say thanks for all your help. There were certainly high and low points on the this week long struggle but you have helped me to end on a high point. Rather than having an 80 euro brick I now have something that looks like it will run open cpn.