Mele armhf 3.0.57+ / 3.0.42+ – kernel update

guillaume

After many requests (sorry for the delay), you will find here new kernel updates for both images (debian headless and ubuntu w/GFX).

UPDATE 5 – Apr 4 2013
GFX update – Updated mali driver + some modules like UVC.

1G:
linux-armhf-gfx-1G-3.0.42-6.tar.bz2 (MD5: 8004415caaf82fd989b8f04cea441deb)
linux-armhf-gfx-1G-3.0.42-6.tar.gz (MD5: 5efbd84a3ebece09b13c353a76ebcf88)

Normal (512M):
linux-armhf-gfx-3.0.42-6.tar.bz2 (MD5: 53c39f9e67061f78975755ef893b83fe)
linux-armhf-gfx-3.0.42-6.tar.gz (MD5: 5abd088b05f19d5c5e25dd8a1ca718a5)
/UPDATE 5

UPDATE 4 – Feb 10 2013
Update only for 1G. Only the cmdline is modified for nand install.

linux-armhf-gfx-1G-3.0.42-5.tar.bz2 (MD5: a9643363452255cf3a11c8d54a467e47)
linux-armhf-gfx-1G-3.0.42-5.tar.gz (MD5: d178e04876280701bf060356a8bcbc54)
/UPDATE 4

UPDATE 3 – Feb 8 2013
This update is only GFX – speedy mali running at full speed with latest mali drivers (and all other things still working like sata).
Same ‘make ‘ as previous.
This kernel runs xbmc ~without~ lag on 720p.
Mali is at full speed (400MHz).

Ensure you have this in your /etc/X11/xorg.conf

Section "ServerFlags"
...
Option "AIGLX" "false"
Option "DRI2" "true"
...
EndSection

linux-armhf-gfx-3.0.42-4.tar.bz2 (MD5: 5fecada9bb36fa6cbce0a911b4476b18)
linux-armhf-gfx-3.0.42-4.tar.gz (MD5: 5fecada9bb36fa6cbce0a911b4476b18)
linux-armhf-gfx-1G-3.0.42-4.tar.bz2 (MD5: fa53c74073fe39cd5eeb5245a6ff2837)
linux-armhf-gfx-1G-3.0.42-4.tar.gz (MD5: b3a41d3e6e5605ef8fddf1e1b90190ef)
/UPDATE 3

UPDATE 2
New release with the correct iptab stuff. Are working fine for me.
One 1G version (forced kernel cmdline) for nand install.

a Makefile is included to install it faster.
For nand install from a nand running device:

# cd <decompressed_dir>
# make nand
# sync

For sdcard install from a sdcard running device:

# cd <decompressed_dir>
# make sdcard
# sync

linux-armhf-gfx-3.0.42-3.tar.bz2 (MD5: a8fcc6b6ac037901e25faba746a4e2ac)
linux-armhf-gfx-3.0.42-3.tar.gz (MD5: c8d60b8ebc2aed61abc56a947c349191)
linux-armhf-gfx-1G-3.0.42-3.tar.bz2 (MD5: 478f860499ae5e100e6e61f26ffc86dc)
linux-armhf-gfx-1G-3.0.42-3.tar.gz (MD5: 88bec841e1fc3f0c8b8fd6347ff6b1a0)

linux-armhf-headless-3.0.42-3.tar.bz2 (MD5: 4088e81c5035b12635220bd408474db6)
linux-armhf-headless-3.0.42-3.tar.gz (MD5: efe8f0adaa9797ad90da37586291dcc4)
linux-armhf-headless-1G-3.0.42-3.tar.bz2 (MD5: a3f2c2a706700b76983eb16fc4b957dc)
linux-armhf-headless-1G-3.0.42-3.tar.gz (MD5: 158d70c8b0f05cbc3b2c6002058843a2)
/UPDATE 2

UPDATE
After some sata problem with 3.0.57+, using this release seems useless against previous 3.0.42+.
So, here are an updated 3.0.42+ version which is working with sata:
linux-armhf-gfx-3.0.42-2.tar.bz2 (MD5: b0ef9e9d75523d9da4fbf36052c6ec39)
linux-armhf-gfx-3.0.42-2.tar.gz (MD5: e48dd1aa78de75e6dadf9540a7a2ab86)
linux-armhf-headless-3.0.42-2.tar.bz2 (MD5: 06aa3ca83aea0e8eab81c6c033011a24)
linux-armhf-headless-3.0.42-2.tar.gz (MD5: 9eaa2a235fd38c58bcb9f4af39ec6e48)
/UPDATE

Linux 3.0.57+ armhf w/GFX support:

Linux 3.0.57+ armhf headless:

These releases include:

  • the needed patches to make them work properly on all version of mele (should be).
  • the removal of the memory limitation for >512M version of mele (should work… i don’t have one 1G to test).
  • All modules … i have put everything that could compile 😉
  • the config file

This release comes from the freshly unified git linux-sunxi. I have just patched the arm clock and wemac driver according to the previous 3.0.42+.

I have make the choice to provide only the kernel with modules here and not reedit full images. It was useless.
To install one of this kernel, you have to do something like that:

Mele running on SDCARD (MMC):

$ su -
# cd /tmp
# wget -nd http://guillaumeplayground.net/share/linux-armhf-gfx-3.0.57.tar.gz
# tar xzf linux-armhf-gfx-3.0.57.tar.gz
# cd linux-armhf-*
# mount /dev/mmcblk0p1 /mnt
# cp p1/uImage /mnt/
# umount /mnt
# cp -a p2/lib /
# chown root:root -R /lib/modules

Mele running on NAND:

$ su -
# cd /tmp
# wget -nd http://guillaumeplayground.net/share/linux-armhf-gfx-3.0.57.tar.gz
# tar xzf linux-armhf-gfx-3.0.57.tar.gz
# cd linux-armhf-*
# mount /dev/nanda /mnt
# cp p1/uImage /mnt/linux/
# umount /mnt
# cp -a p2/lib /
# chown root:root -R /lib/modules

Have fun.
Bye!

137 thoughts on “Mele armhf 3.0.57+ / 3.0.42+ – kernel update”

  • Many many thanks for this good work on Sunday
    I’ve been waiting all excited for the update.

    I will test it immediately.

    Reply

  • Thanks for taking the time and providing the kernel modules! For my purpose I dont need sata.. so sticking with linux-armhf-gfx-3.0.57

    Reply

      • Hi Guillame.

        Thank you, I have followed your suggestion and now if I boot from the SD card it recognize the 1GB:

        root@mele:~# free -m
        total used free shared buffers cached
        Mem: 992 26 965 0 3 12
        -/+ buffers/cache: 11 980
        Swap: 0 0 0

        However, I have a Debian already installed in my Mele 2000G, should I reinstall all again or what should I do in order the nanda installation recognize the 1GB.

        Reply

        • For nanda boot, it’s tricky because this kernel has the boot params inside and the vanilla android uboot does not allow extended parameters.

          It’s quite stupid but the only simple way could be to release a 1Go version (only changing the boot parameters).
          I will make one with the iptables update.

          Reply

          • Is there a simple way to change the boot parms for NAND install?

            i have the cubieboard 1gig memory. It’s only allocating 512megs.

          • Let me please know once you’ll release kernel with mem=1024M parameter…I can’t wait to use my laptop with full RAM!

            Thank you

          • Hi Guillaume.

            Just to let you know that the version you release with the 1GB kernel support is working so far in my Mele 2000G.

            Thank you very much for your hard work.

          • Sadly no… mali.ko can’t be loaded, because it’s address. Or…it can be laoded and system boots, but few seconds after insmod, system freezes, writing to logs the same message about bad address. it seems like mali.h has declared different mem block mali.ko uses, then my combo has defined in it’s script.bin.

  • Dear Guillaume,

    Thank you for your updates.

    I have tried to update the headless and gfx version, but the USB is still not working.

    Do I have to place the config file in a folder?

    Reply

    • The config is the kernel config – no use unless you want to compile a kernel.

      3.0.42:
      USB, SATA, WEMAC, WIFI, GFX are working. Seems only iptables to be missing. I will release a new one with iptables working.

      3.0.57:
      USB, WEMAC, WIFI, GFX are working. I don’t get sata working… a clock problem… always a clock problem with the mele.

      Reply

  • It´s possible that after upgrading the kernel with linux-armhf-headless-3.0.42-2, mele doen´t boot? I´m going to flash again and restore a backup, but I don´t know if I must try again with the kernel upgrade.

    Thanks.

    Reply

    • It should work. The 3.0.42-2 is just the same kernel as previous with more modules (i will make one with a correct iptables this week-end).
      What kind of mele do you have?

      To restore:
      If you are running it on a SDCard, you can simply mount the SDCard and reinstall the previous kernel (Partition1: uImage, Partition2: /lib/modules)
      If you are running it on nand, you can boot from the image on the SDCard and restore the previous kernel and modules.

      Reply

  • Thanks for your answers and time. My Mele is a A2000 with 512 Mb of Ram, and it works perfectly with your debian image. I did an apt-get update, apt-get upgrade, make a backup with tar, and then, I tried upgrade the kernel with your instructions. I´ve got the Mele running in the nand, but with the second try, is the same situation. I can´t connect throught ssh, and if I sniff the net, I don´t see anything from Mele. It´s configured with static ip, and the router has enabled dhcp.

    Thanks

    Reply

    • Same here. When I do nand installation mele doesn’t boot completely. Ping works, but no video output and no ssh connection.

      Reply

  • If I boot from sdcard, mount the nand partition, and take a look to /mnt/var/log/dmesg, I can see this error:

    [ 10.640000] dev_addr= 00:51:25:52:8d:3e
    [ 10.680000] bDriverStopped:1, bSurpriseRemoved:0, bup:0, hw_init_completed:0
    [ 13.380000] wemac wemac.0: WARNING: no IRQ resource flags set.
    [ 13.390000] wemac wemac.0: eth0: link down
    [ 13.390000] ADDRCONF(NETDEV_UP): eth0: link is not ready

    Reply

    • This is odd.
      It’s not able to load the wemac?
      The 3.0.42-2 is the same version as the previous one.
      Perhaps you could try to configure the wifi.

      Did you try the GFX version or the headless?
      Which image are you using?
      I will check it and make an update this week-end.

      Reply

    • Same here. When I do nand installation mele doesn’t boot completely. Ping works, but no video output and no ssh connection.

      Reply

  • Well, I´ve restore with your method, reinstalling the previous kernel. I don´t know what can be the problem. Thanks anyway.

    Reply

  • Just wondering: If this is the same kernel as the previous one (the only difference being the amount of kernel modules supplied), could we just copy the modules and use the existing kernel in Debian headless?

    Reply

  • Is there a working kernel supporting 1024MB RAM? I have tried yours 3.0.57+ version, and it works, except the RAM. I need my RAM badly, because I am using your image on A10 Boxchip SoftwinerEvbV13 laptop…and it’s desktop is slow as hell with default 512MB RAM.

    Reply

  • BTW, is there a way to use full potencial of mali gpu?
    I have tried to load mali driver, and mali_drm loads fine, but mali module tries to use bad address.

    FATAL: Error inserting mali (/lib/modules/3.0.42+/kernel/drivers/gpu/mali/mali/mali.ko): Bad address

    Reply

    • Hum… probably due to the ram extend.
      Could you try to change the kernel cmdline with a ‘mem=960M’ and see if the mali driver loads?
      I have no 1Go version to try on.

      Reply

      • I have used your new release, and it works fine. My uboot ignores params file, so it was sufficient to copy your uImage and modules, and everything works, except for MALI, which is slow as hell with fb.

        Reply

  • I am sorry to spam your devel thread for mele with things needed mostly for different mobo’s, but we are talking about A10, right? It seems that mali driver is compiled with framebuffer support only. Is there a way, or can I ask you for this one, to compile kernel with X11 support with mali.ko module? With valid address, so the module can be loaded and used? Mali is very capable gpu, and it’s poor performance right now is sad…

    I would compile it myself, but I don’t have kernel sources you’ve built the kernel from (with patches).

    It would be nice to not call this “mele” distro, but generic for A10 chipboards. And thus, have a full mali support would be nice.

    Reply

  • Loaded the new 1G kernel on my cubieboard. It loaded fine but the memory is still at 512. I’m running the image off the sdcard, so I don’t know if that is the problem or not.

    I also included another line in rc.local for SSD devices. I changed the i/o scheduler from cfq to noop. Helps for speed.

    Dwaine

    Reply

    • I did the nand copy and the image running off the nand uses the full 1g of memory.

      Now if I could figure out how to get my updated version of boot.scr to actually load with the mem=1024M parameter. Life would be good.

      So far your image and kernel are the best option for cubieboards.

      Here are my changes to rc.local if anyone cares….

      cpufreq-set -g performance

      echo performance > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
      echo 1008000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
      echo 1008000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq

      echo noop > /sys/block/mmcblk0/queue/scheduler

      Dwaine

      Reply

      • For the ubuntu image, you can simply do that to remove the ondemand governor:
        rm -f /etc/rc2.d/*ondemand

        In my image, you already have this:
        echo “cpufreq-set -g performance” >> /etc/rc.local

        For the scheduler, you’re right too. noop for mmc and nand.
        Bye.

        Reply

  • Hello again,
    after two days of fiddling with my new toy, I have some thoughts I want to share, if you don’t mind.

    I have Boxchip A10 SoftWinerEvbV13 laptop, which is just another copy of Allwinner A10. This model is equipped with 3x USB, 10” screen, physical keyboard, eth0 and wlan0 interfaces, SD reader, touchpad and MALI-400 GPU.

    It has also four LEDs, first for ON/OFF state (white), second for charge state (red/green), third for Wireless/Boot state (blue), and fourth for CAPS (green).

    When I bought it, it had Android 4.0.3 installed, kernel built by some guy named “crane”.

    After a long time searching the net, I have realized that I have just some unknown clone of A10, and seeing it’s internals made my feeling real.

    So I knew about Ubuntu images, but it seemed like waste of free space, because it is SD based only. I wanted to run from NAND. So here I am.

    I must thank author of this distro image and it’s kernel – Guillaume

    My Setup:
    – booting from original Android’s uBoot, system in NAND
    – to free some space, I have moved /usr to second partition of my SD, which is mounted as /usr
    – first partition of SD is mounted as /home, third works as swap space
    – kernel: linux-armhf-gfx-1G-3.0.42-3 with it’s modules
    – desktop: kde4

    What Works:
    – audio (mpg123 needs to be recompiled with –with-cpu=generic_fpu parametter to make mp3 with fpu codec play), and need to switch default audio device in mixer
    – video (kernel even knows we have strong hw, /dev/dri/card0 and /dev/dri/card1 does exist – to have display work, you have to copy script.bin from original android image to your boot partition)
    – keyboard
    – display
    – touchpad
    – wireless
    – ethernet
    – various devices (USB 3G card, wireless mouse, etc)

    What does not work:
    – advanced 3D: mali kernel module can’t be loaded, throws (FATAL: Error inserting mali (/lib/modules/3.0.42+/kernel/drivers/gpu/mali/mali/mali.ko): Bad address)
    – due to desktop running in the framebuffer, it is little bit slower. Glxgears shows 35fps (working mali module should give ~140fps).
    – compozition is realized via Xrender. Some animations does not work.
    – audio has bad default device selected. It should be set to sun4icodec by default (ubuntu mechanics….)
    – due to missing some attributes of hw accelerated graphics (missing mostly due to framebuffer), plasma-desktop may misbehave (artefacts, or some buttons are deformed and do not react upon click).
    – kernel does not see battery…it can detect charging and discharging states, but nothing more, no battery % or so.
    – kernel has issues with sleep mode, it does not work mostly, it just switch off/on display.

    Well, taking the fact that this was “mele” oriented thread/devel forum/site, great job!
    I must note your PIMP_MY_MELE tool, which is great for NAND write in two stepps. But it’s VGA capabilities should be extended for more outputs. I know – mele was not supposed to have fancy graphics and 1GB of RAM, or even touchpad. So…I am very pleased. I want to thank you. And by writing this here, maybe help somebody else with the same laptop 🙂

    For me, it is almost ok. I can live with it, because I need it as portable terminal mostly…but have a chance to play some simple game sometime would not be bad… 😉

    So again, thank you

    Reply

    • Virnik,

      Thank you very much for your comment.
      I’m just a user/player/hacker/developer like you on this platform.
      If this website and my little work helped someone to play and learn with funny hardware like the A10, this job is done 😉
      The Allwinner/Mali support is just too low.
      Now, we have a buggy simple GLX and a seemly working DRI2.

      I think we have to take the A10 like it is: a great performing platform for money, an early vision of a real mainstream soc… which will probably be more open and supported.
      If your hardware does the job for you now, this is probably it.

      For me, it is, i have a super-lowcost, low-power:
      – 32x1GHz http server cluster with sata ssd
      – 1 bit-perfect audio player
      – 2 gateways/firewalls/router with vlan

      All running linux, all running without failure.

      So many great hardware in the past two decades were born dead because of the closed software they were hosting, perhaps we could thank android to become the new standard pseudo-opened platform… perhaps.
      We will see in the next years how it goes 😉
      One more time, thank you and thanks everyone for the time spend on playing on this platform.

      Bye.

      Reply

      • Hello again, I have been doing some tests, made my own builds, and it shown itself useless. I have my kernel booting, but it seems that I am something missing…

        Can you please take a look on the kernel again? I have tried few nightly builds of kernel from http://dl.linux-sunxi.org/nightly/
        and those works with mali kernel module just fine, even it’s performance is much more higher then it is in your kernel. Sadly, they do not have 1024MB RAM support. I understand that I can’t have whole RAM for userspace, because some have to be left for GPU and some for ramdisk. But anything is better then live with working mali, but only 340MB RAM at least.

        Will you be so kind and try another build? I don’t care about sata, which is still buggy as it seems, I need only desktop laptop for my child, and myself in case of some network device’s failure (portable terminal with ethernet and wifi).

        In the hwpack for mele, there are working mali GL ES libraries, which works fine. They are missing in your image, but I suppose that you can include them in standalone hwpack (libs and modules).

        Can it be done?
        And specify your paypal please, I would like to send you some cash for your time.

        Thank you very much.

        Reply

        • Hi,

          But the mali driver is working with my kernel, right?
          I get enough speed on mine.

          To activate DRI2, you need this in your /etc/X11/xorg.conf:

          Section "ServerFlags"
          Option "AIGLX" "false"
          EndSection

          So you get this in /var/log/Xorg.0.log:

          [261373.268] Loading extension GLX
          [261373.268] (II) LoadModule: "mali"
          [261373.269] (II) Loading /usr/lib/xorg/modules/drivers/mali_drv.so
          [261373.269] (II) Module mali: vendor="X.Org Foundation"
          ...
          [261373.282] (II) MALI(0): DRI2 version: 1.2
          [261373.282] (II) MALI(0): [DRI2] Setup complete
          [261373.282] (II) MALI(0): [DRI2] DRI driver: Mali DRI2
          ...
          [261373.284] (WW) MALI(0): Initializing EXA Driver!
          [261373.285] (II) EXA(0): Driver allocated offscreen pixmaps
          [261373.285] (II) EXA(0): Driver registered support for the following operations:
          [261373.285] (II) Solid
          [261373.285] (II) Copy
          [261373.285] (II) Composite (RENDER acceleration)

          Verify your a10 is running at full speed:

          # grep BogoMIPS /proc/cpuinfo
          BogoMIPS : 1001.88

          If not:

          rm -f /etc/rc2.d/*ondemand
          echo "cpufreq-set -g performance" >> /etc/rc.local
          cpufreq-set -g performance

          The hwpack contains a kernel, modules and libs.
          Is it just the libs or the mali module which change the speed?
          I don’t see difference with the libs from my mele.

          You could try my kernel with these libs? (the ones from the hwpack directory a10-bin-backup/lib)

          7zr x mele-a1000-vga_hwpack*.7z
          cd mele-a1000-vga_hwpack...
          cp -a rootfs/a10-bin-backup/lib/* /lib/
          ldconfig

          For the module, i have the vanilla one.
          If you still have slow mali, i will take a look at the changes of the kernel from the sunxi tree.

          For the sata, my precious, i need it 😉 the mele is one of the rare platform with and it’s working pretty well. I will be a shame to release a kernel without it.

          For paypal, i’m making this for the fun but if you insist, it will be a pleasure 😉 gfoustar@gmail.com

          Reply

          • Thanks for your reply. I am doing this for fun mostly, but I can’t deny practical thing in it. For the CPU, I am running on performance governor, I have min/max cpu step set, so I am really running 1GHz.

            For the RAM, system reports something around 940MB of RAM, which is perfectly fine (something in ramdisk, and something for GPU, if I count right).

            But graphic performance is poor, at best. Display is really choppy, I thought that composition should work without any choppiness.

            Regarding mali kernel module, even X says it’s loading it, I do not trust it. It seems that it runs mali, but mali module is not loaded, and can’t be loaded- so it is framebuffer with mali_drm and DRI2 only. It complains about bad address. So for example glxgears shows only 30fps max. And I saw the same chipset with the same ubuntu version on youtube, where they had something about 140fps max from glxgears. Video of any kind is unplayable, but mali gpu should render 720p movies without issue.

            I don’t have my A10 laptop with me at work, so I will try new hwpack later today, maybe in the night.

            For paypal, you should have $10 there from virnik@gmail.com, so have a drink 😉

            And thanks

        • Hi,

          From xbmc tryout, i have something working ~well~ now with mali and sata.

          I will post a new kernel, libs and perhaps a new ubuntu image.

          And thanks for the drink! 😉

          Reply

          • You are glorious… 😉
            You know, combination me having a lot of work, newborn child and lazyness is very sad…so you have made my day, thank you very much.

            And few more tips:
            1) maybe giving eth0 static IP will help a lot of people. Not everybody has linux dhcp server at home like me or you (or access to it’s logs).
            2) releasing complete X based image would be fine too, especially for lames (well, I prefer KDE, so I have working K desktop now, but it depends on personal feeling)
            3) changing hostname from ‘mele’ to ‘ubuntu_armhf’ (or anything else) may help too. Not everybody here uses mele, the identical chipset counts, not manufacturer/make/model. but it is just cosmetics
            4) releasing kernel/libs/modules in one separate package like you do is definitelly good step, especially for updates
            5) PIMP_MY_MELE script should understand DVI and it’s resolution…I have tried manual modification of script.bin, which worked, but framebuffer flickered from time to time then…so I am back to original android script.bin

            Keep up with good work.
            And for the drink, I glad you enjoy it… 😉

          • I have something.
            Reading through diagnostic messages, while laboring in the shell (all modules commented out), I left one session to log actual kernel messages and store them to the sd

            Now I have this:
            [ 277.500000] mali: use config clk_div 3
            [ 277.500000] mali: clk_div 3
            [ 277.510000] Mali: mali clock set completed, clock is 400000000 Mhz
            [ 277.520000] Mali: ERR: drivers/gpu/mali/mali/common/mali_kernel_core.c
            [ 277.530000] mali_parse_config_memory() 581
            [ 277.530000] Failed to register MEMORY

            Whole system frozen afterwards. So it is definitelly mali module, and maybe something with RAM.

            root@PinkPanther:~# cat /proc/cmdline
            root=/dev/nandb mem=1024M rootwait console=ttyS0,115200

            I Think that you have to leave some RAM for GPU, which mali.ko uses. Can you please rebuild 1G version of your kernel with this cmdline changed? I think that having ‘mem=960M’ there may mali.ko work. But I haven’t seen sources, what the kernel config is. I think that some time back when I was compiling kernels for all my pc’s, there was some option to specify gpu pre-allocated ram, and then what is the full system ram.. I mean during config.

          • Strange is, that I already don’t have 1024MB of RAM. Isn’t there some kind of missconfigurations, relict from build for 512MB of RAM?

            root@PinkPanther:~# free -m
            total used free shared buffers cached
            Mem: 879 175 704 0 19 103
            -/+ buffers/cache: 51 828
            Swap: 1019 0 1019

        • Here we go!
          I have made a new update available here.

          This kernel have the lastest mali driver and is running quite nicely for me so far.
          Able to run xbmc also.

          Bye.

          Reply

          • Thanks for your submission, but that kernel won’t boot. It simply lit display, but nothing will show up. I have searched logs, and they are created upon booting, but are empty. It seems that some module loaded during startup segfaults or make kernel panic.
            As it seems…I have no screen during boot, because of plymouth.

            Can you add “no plymouth” to the kernel cmdline?
            I don’t have any logs, because even logs are created during boot, they are all empty. I suppose mali moduel behave wrong…

            Previous version worked fine.

          • Odd.
            It’s booting on my two mele here.
            So i can not test easily.

            You can try to increase the mali_clkdiv in your script.bin like this.
            It will put back the previous clock speed on mali (320Mhz) but with the new drivers. The other 3.0.57 (and the one from the image i tried) is running at 320 but i’m able to run at 400Mhz with my 3.0.42.


            # mount /dev/mmcblk0p1 /mnt (or nanda)
            # /usr/local/sunxi-tools/bin2fex /mnt/script.bin /mnt/script.tmp.fex
            # sed -e 's/^mali_clkdiv.*/mali_clkdiv = 4/' /mnt/script.tmp.fex > /mnt/script.tmp2.fex
            # /usr/local/sunxi-tools/fex2bin /mnt/script.tmp2.fex /mnt/script.bin

          • OK, will try in one hour…I have to let my son sleep…so it will take a while. Thanks for the tip, I will try what it will do.

          • So far so same…System starts booting, but kernel stops few seconds afterwards. No logs.

            If you are willing to investigate further, I can send you my script.bin, if it may help. What was biggest difference between linux-armhf-gfx-1G-3.0.42-4 and linux-armhf-gfx-1G-3.0.42-3. Older works fine except for mali.ko, while new one does not boot.

          • Difference with the kernel is only the latest kernel mali/ump modules.

            In the Makefile of the package:
            – Install latest Mali library (deployed in /lib/).
            – Changes script.bin mali_clkdiv to 3.

            That’s all for me i think.

            You could try to remove from the script.fex these lines:

            Lines 12-13
            -- pll4_freq = 960
            -- pll6_freq = 1200

            restore mali_clkdiv = 3
            and regenerate scrip.bin

            Comment these lines in /etc/modules:

            # lcd
            # hdmi
            # ump
            # disp
            # mali
            # mali_drm

            Boot the kernel and activate one by one each module with modprobe … and see what happen and which one is blocking.

          • I did as you said, and it is really mali module.
            If I’ll boot without it, everything works. If I’ll try to load it manually, I get this:

            root@PinkPanther:~# modprobe mali
            FATAL: Error inserting mali (/lib/modules/3.0.42/kernel/drivers/gpu/mali/mali/mali.ko): Bad address

          • Damnit, my previous replies has ended up in upper thread. Nevermind. I may have a solution.
            Take a look here:
            http://forums.arm.com/index.php?/topic/15365-how-could-i-make-sure-the-driver-work-success/

            Starting with post #11, it starts to be interesting. Simply said, for A10 chips inside tablets and laptops, there are different RAM blocks used for fb and mali module.

            I’ll bet that there are at least two versions, first is mele, and similiar devices, and second are laptops/tablets allwinner.

  • One last thing – audio decoder issue can be solved easier, by switching from mpg123 to ffmpeg in every app you can. Playback works fine then

    Reply

  • Your SoC cluster is impressing…
    Can I ask you one last question? Can you compile, or will you give me instructions for it – mali driver again? I mean so that kernel creates /dev/mali, which is needed for mali libs compilation (Like you said, it is a toy. I am not developer, I am just “bender” – or cracker. I can do some things, but not all. I know linux well, but all this ARM is very new to me).

    Do you have some paypal account, so I can donate few bucks for your effort?

    Reply

  • Hello. Even with the latest update (UPDATE 2), I can´t connect to the device via Ethernet. I tried to do the update, either manually, as with makefile. The same error in wemac. I’ll enable wifi before reflash the kernel one more time. Thank you for your work and your effort.

    Reply

    • After configuring wifi, now I can access via ssh to Mele. Ethernet is down and leds is off, but now I can investigate about this. With the kernel update, finally, I made the mpd works, but something curious happens when I launch aplay -l command, I can see the hdmi soun device, when my system is Debian headless:

      aplay -l
      **** List of PLAYBACK Hardware Devices ****
      card 0: sun4icodec [sun4i-CODEC], device 0: M1 PCM [sun4i PCM]
      Subdevices: 1/1
      Subdevice #0: subdevice #0
      card 1: sun4isndhdmi [sun4i-sndhdmi], device 0: SUN4I-HDMIAUDIO sndhdmi-0 []
      Subdevices: 1/1
      Subdevice #0: subdevice #0
      card 2: Device [USB Sound Device], device 0: USB Audio [USB Audio]
      Subdevices: 1/1
      Subdevice #0: subdevice #0

      I´ve used the thir device, and it´s sounds with headphoen output!!! Many, many thanks for everything!!!

      Reply

      • Oh nice.
        Perhaps you took the kernel -gfx ?
        Also, you can disable the hdmi by commenting out these lines in /etc/modules:

        # lcd
        # hdmi
        # ump
        # disp
        # mali
        # mali_drm

        Which usb audio are you using?
        I confirm mpd + mele is the perfect combo (without resampling).
        Mine just rocks 😉

        Reply

  • It´s possible that having chosen the wrong kernel (gfx), this is the reason for my ethernet problems? My usb audio it´s a Sweex 7.1, a C-Media CM6206 based card. A cheap sound card, compared with yours :D, but doesn´t sound bad. I have to modify the configuration, because I can´t control de volume with android mpd client. Thanks.

    Reply

    • I don’t know this dac.
      Mine is not too expensive… compared to a lot of overpriced dac.
      But compared to the $70 mele, it is 😉

      For the volume control, if you have a hardware one, you should see it appear with ‘amixer scontrols’.
      After, you just have to take what’s between the simple quotes and set it in your /etc/mpd.conf.


      # amixer scontrols
      Simple mixer control 'MICROMEGA Clock Selector',0
      Simple mixer control 'MICROMEGA Clock Selector',1

      /etc/mpd.conf

      audio_output {
      type “alsa”
      …
      mixer_control “MICROMEGA Clock Selector”
      …
      }

      Reply

      • Is there a way to check which address mali module needs? I suppose it’s from script.bin, or it’s decompiled script.fex. Because it seems that mali kernel won’t load if I have mem=1024 on kernel cmdline… So I suppose this value will have to be reduced….

        Reply

        • mali takes 64Mo.
          But the kernel “should” reduce by itself the total amount when loading mali…

          can you try mem=960M from sdcard ?
          If it works, i will make a new update with this cmdline.

          Reply

    • If it boots from sdcard, it should boot properly from nand.
      Are you forcing some params in boot.scr with sdcard boot?.. like the mem for example? (check your available mem in sdcard boot)

      It could be the same report as Virnik if you have <512M with sdcard boot.

      With this kernel, do you feel the speed increase with xorg?

      Reply

      • I just upgraded the kernel to v4 from the sdcard to the NAND from the PIMP_MY_MELE script. The cubieboard has 1gig of memory so I used the 1G version of the kernel. The previous upgraded kernels including 3.0.57 upgrade boot fine after upgrade and install to NAND.

        Its funny, when the run the 1G version of the kernel from the sdcard, it only sets up the memory to be 512megs. If I install the kernel from the sdcard to NAND it boots with 1G.

        Off the sdcard, xorg loads fine. Have not noticed a real speed increase. I did notice that in your instructions, it list

        Section “ServerFlags”
        …
        Option “AIGLX” “false”
        Option “DRI2” “true”
        …
        EndSection

        But in your Ubuntu image there is no ServerFlags section. The options are in the “Device” section. Also noticed that there are two options “DRI2” “True” and no option “AIGLX” “false”. Xorg for me does not load AIGLX, just DRI2.

        Reply

        • Ok, if i undestand correctly, Virnik guess was good. It doesn’t work with mem=1024M.
          This stupid kernel have an hardcoded bypass to allocate 64 from 512 but not over…
          I will post a new one 1G with mem=960M tonight.

          For AIGLX, it must be desactivate. Mali is GLX.
          Xorg by default loads AIGLX, if you don’t ask xorg to bypass AIGLX, it will never load GLX for the Mali.

          You need to add this section ServerFlags in your xorg.conf.

          Reply

          • You are my savior! Thank you. I inspected the problem, even tried myself to compile working uImage..it works, except for USB devices and sound, so I suppose I have missed something. And I don’t feel like I’ll start compiling hid and usbcore modules, not talking about ALSA (It took too long few years back on my old PC). But it loaded mali module fine, and I had pretty performance increase in glxgears. I had something around 142fps, which is pretty fine compared to my actual 28~36fps. So again, I welcome your goodness and will to do perfect kernel for everybody.

            Thank you.

          • It’s available on this page.

            Please tell me if it’s ok – and if GFX is running fast (enough).
            It’s the new mali driver (not the one from the sources i shared in another post).

            I will update the ubuntu/debian image if all is fine (integrate the 1G/512M switch in PimpMyMele, package a xbmc deb also)

            Bye.

          • Outch…

            I don’t know how you get a 140fps with glxgears.
            On my mele, it tops at a stupid 36fps.

            Which kernel did you compile ?
            Could you show me your config ? (content of /lib/?, is it the original xorg from my ubuntu?, xorg.conf?)

            I don’t get it because i don’t know much about all this gfx stuff.
            The only thing is eglgears should be a more real benchmark than glxgears.

      • I redownloaded the v4 1G kernel again. Tried the upgrade on the sdcard (which worked fine) and then install from sdcard to NAND. The cubieboard boots, but get suck somewhere. Just a blacken screen.

        I do know that some of the booting gets completed. My router does see the wifi usb dongle, but does not assign a IP address. So the driver loaded, but the network services does not fire up I guess.

        I would also love to see in your PIMP_MY_MELE script the option of copying from NAND back to sdcard. That why I could backup my NAND install and if anything goes wrong, just pop in the sdcard and use your PIMP_MY_MELE script to install the backup copy to NAND again.

        Reply

        • You are stuck at mali.ko load. The load itself fails, kernel refuses to map already allocated memory block. But five seconds later, it’s memory print spoils whole memory block, corrupting it and thus making kernel Ooops!. Boot with recovery SD (did you do one, right), and mount /dev/nandb to /mnt
          then in /mnt/etc/modules file, comment out ‘mali’ module. Unmount /mnt, and reboot. It will work like it did with older kernel release, but composition will be run over xrender, and will be basic only. So no mali advantage.

          Reply

  • I have started to celebrate too fast. It seems that mem= parametter is not enough. Problem is in the mali.ko source, which needs to load to specific address.

    It can be seen in the script.bin, respectively in it’s decompiled form, which address it uses.

    I still do have this msg:
    root@PinkPanther:~# modprobe mali
    FATAL: Error inserting mali (/lib/modules/3.0.42/kernel/drivers/gpu/mali/mali/mali.ko): Bad address
    With following system freeze.

    Here is my script.fex:
    http://pastebin.com/HMHe6vx8

    I suppose that there is the right address, and I bet my bare foot, that The others with the same symptoms like I do, will have the same. At least regarding Mali. From what I have found on the net, it seems that there are two separate board versions, which differs only in little details…Mali address is one of them. At least it was also said in the links to the threads I’ve posted earlier.

    Dear Guillaume, you have done enough. I thank you for your work and will to help, but I don’t want to trade you on for this more over it. Fixning this will require you to relese two seperate kernels for two mainboards, each with different memory address in mali.h, which is little bit too much for me to ask. I am grateful that I have at least working desktop, thanks to you, which I can use as portable terminal. Sadly, no games on it, but that does not matter so much.

    Keeping in mind that this all started as your hobby for your mele mobo, it seems heroic that you have made owners of different mobos with the same arch happy 😉

    Reply

  • Thankyou Guillaume!!

    I am still using the headless version you released in november, it’s working so well I’m a bit scared to update…ehehe

    Did you experience big improvements that should lead me to take the risk?

    Reply

  • Dear Guillaume,

    Thanks again for sharing your work!
    I just the new patch on my A2000G. Here is exactly what I did (on my desktop):
    sudo dd if=debian_wheezy_armhf_v1_mele.img of=/dev/sdg bs=1M

    then I booted the A2000G on the sdcard and
    root@mele:/ free -m
    total used free shared buffers cached
    Mem: 484 96 388 0 3 80
    -/+ buffers/cache: 11 472
    Swap: 0 0 0

    root@mele:~# cd /tmp
    root@mele:/tmp# wget -nd http://guillaumeplayground.net/share/linux-armhf-gfx-1G-3.0.42-5.tar.gz
    root@mele:/tmp# tar xzf linux-armhf-gfx-1G-3.0.42-5.tar.gz
    root@mele:/tmp# cd linux-armhf-gfx-1G-3.0.42-5
    root@mele:/tmp/linux-armhf-gfx-1G-3.0.42-5# make sdcard
    root@mele:/tmp/linux-armhf-gfx-1G-3.0.42-5# reboot -n

    After reboot :
    root@mele:~# free -m
    total used free shared buffers cached
    Mem: 308 25 282 0 3 13
    -/+ buffers/cache: 9 299
    Swap: 0 0 0

    I’m still trying to get the default script.bin from the nand.

    Best regards,

    Nicolas

    Reply

    • Hi,

      It’s ok with the image and the kernel like this.
      Please try this command on the mele:

      dmesg | grep 'command line'

      You will see, with your sdcard boot, something with a mem=XXX
      It is this option to change to mem=960M.
      If you install to nand, it will work without changes… because there’s no boot.scr.

      For the sdcard, you need to edit the (/dev/mmcblk0p1)/boot.scr
      On the mele, mount your sdcard p1 and proceed like explained on:
      http://rhombus-tech.net/allwinner_a10/hacking_the_mele_a1000/Building_Debian_From_Source_Code_for_Mele/
      Here: “Create a boot.cmd file with the following contents:…”

      You will have to change the “extra” env in the example with:

      setenv extra 'mem=960M rootwait'

      Build boot.scr, umount and reboot. It will do the trick.

      Reply

  • Hi Guillaume,

    Here the dmesg output:
    root@mele:~# dmesg | grep ‘command line’
    [ 0.000000] Kernel command line: console=ttyS0,115200 root=/dev/mmcblk0p2 loglevel=7 panic=10 rootwait

    then I create a text file, boot.cmd with :
    setenv root ‘/dev/mmcblk0p2’
    setenv panicarg ‘panic=10’
    setenv extra ‘mem=960M rootwait’
    setenv loglevel ‘8’
    setenv setargs ‘setenv bootargs console=${console} root=${root} loglevel=${loglevel} ${pan
    icarg} ${extra}’
    setenv kernel ‘uImage’
    setenv boot_mmc ‘fatload mmc 0 0x43000000 script.bin; fatload mmc 0 0x48000000 ${kernel};
    bootm 0x48000000’
    setenv bootcmd ‘run setargs boot_mmc’

    and compiled/copied to the sdcard
    root@mele:~# mkdir /mnt/p1
    root@mele:/mnt/p1# mount /dev/mmcblk0p1 /mnt/p1

    mkimage -A arm -O u-boot -T script -C none -n “boot” -d boot.cmd boot.scr
    sudo cp boot.scr mnt/
    root@mele:~# mkimage -A arm -O u-boot -T script -C none -n “boot” -d boot.cmd boot.scr
    Image Name: boot
    Created: Fri Jan 1 07:07:38 2010
    Image Type: ARM U-Boot Script (uncompressed)
    Data Size: 423 Bytes = 0.41 kB = 0.00 MB
    Load Address: 00000000
    Entry Point: 00000000
    Contents:
    Image 0: 415 Bytes = 0.41 kB = 0.00 MB
    root@mele:~# cp boot.scr /mnt/p1/

    then reboot -n and …
    root@mele:~# free -m
    total used free shared buffers cached
    Mem: 816 28 788 0 3 13
    -/+ buffers/cache: 11 805
    Swap: 0 0 0

    Much better, many thanks !

    About the nand install, I still don’t manage to see the nand partitions on my A2000G (see http://guillaumeplayground.net/mele-a2000-headless-debian-wheezy-armhf-with-nand-install-v1/#comment-616) but I’m still trying!

    Reply

    • Oups a typo! instead of

      and compiled/copied to the sdcard
      root@mele:~# mkdir /mnt/p1
      root@mele:/mnt/p1# mount /dev/mmcblk0p1 /mnt/p1

      mkimage -A arm -O u-boot -T script -C none -n “boot” -d boot.cmd boot.scr
      sudo cp boot.scr mnt/
      root@mele:~# mkimage -A arm -O u-boot -T script -C none -n “boot” -d boot.cmd boot.scr
      Image Name: boot
      Created: Fri Jan 1 07:07:38 2010
      Image Type: ARM U-Boot Script (uncompressed)
      Data Size: 423 Bytes = 0.41 kB = 0.00 MB
      Load Address: 00000000
      Entry Point: 00000000
      Contents:
      Image 0: 415 Bytes = 0.41 kB = 0.00 MB
      root@mele:~# cp boot.scr /mnt/p1/

      one should read:

      and compiled/copied to the sdcard
      root@mele:~# mkdir /mnt/p1
      root@mele:/mnt/p1# mount /dev/mmcblk0p1 /mnt/p1
      root@mele:~# mkimage -A arm -O u-boot -T script -C none -n “boot” -d boot.cmd boot.scr
      Image Name: boot
      Created: Fri Jan 1 07:07:38 2010
      Image Type: ARM U-Boot Script (uncompressed)
      Data Size: 423 Bytes = 0.41 kB = 0.00 MB
      Load Address: 00000000
      Entry Point: 00000000
      Contents:
      Image 0: 415 Bytes = 0.41 kB = 0.00 MB
      root@mele:~# cp boot.scr /mnt/p1/

      Reply

    • Hi Nicolas

      Were you ever able to install on nand? If yes, could you please share how you made it work? I have to say that i’m absolutely new to the Linux world (i’m a Windows guy), but I tought this is a good start for a cheap / low power PC/Server project to run 24/7.

      Thank you in advance for your help

      Reply

  • I have teied you image on my 1G Cubieboard. It work you of the box. The distro seems fast and efficient so I would very much like to use it permanetly – but i’m missing my 1G ram.I only get 306Mb.

    I am using your 1G images.

    I have tried the recipie that Jerome and other described above on my cubieboard. After the mkimage command i reboot but the Cubieboard won’t boot to a state where i can ssh into the unit.

    Unfortunately I have no monitor with HDMI and the cubieboard only have HDMI output so i can’t check the console.

    Is there a recipy for getting access to all the ram on a cubieboard?

    Reply

  • I got my mele up and running but I’m pretty sure the mali module is not functioning right.

    I get less than 1 FPS on XBMC, 1-3 FPS on minetest.

    the mali,mali_drm,disp and hdmi modules are loaded.
    I’m using VGA running at 1280×1024 (just for testing, I’ll be using hdmi -> 1080 later)

    I’ve applied linux-armhf-gfx-3.0.42-4.tar.bz2, on top of the rev3 image.

    thanks for your help

    Reply

      • I had my my processor on a really slow governor. Don’t know why; followed your instructions here and now it’s faster, in general, but not in video.

        glxgears is giving me 38-40 FPS, but says:
        libGL error: Failed to load module Mali DRI2

        Reply

      • I have both kernel 3.0.42 and 3.0.42+, by default booting to 3.0.42.
        I rmmod’d mali, mali_drm and disp_ump and then insmod’d them back, but the 3042+ version.
        mali and mali_drm worked ok, but disp_ump gave error -1, invalid parameters

        I’m starting to think that this might be the problem

        Reply

      • on xorg’s log I get:
        Falling back to old probe method for mali
        VGA arbiter: cannot open kernel arbiter, no multi-card support
        MALI(0): Initializing EXA Driver!
        (those are just the warnings, tell me if you want the full log)

        Reply

    • Great! working quite ok you see.
      I have my own which is a bit more recent one, but it’s working quite the same.
      I will release it if i have time.

      You can run xbmc without Xorg also.
      From the console, just launch:

      /allwinner/xbmc-pvr-binhf/lib/xbmc/xbmc.bin

      enjoy. you have the full speed mali at 400MHz, sata and everything.
      Quite fun for this little box 😉

      Bye.

      Reply

  • Last 3 questions and I promise I’ll stop bothering!

    have you tried overclocking? http://forum.doozan.com/read.php?6,8529,8963 (search for ‘warning’)
    I succesfully OC’d to 1.2 on android.

    Do you know of any other ubuntu repo for arm? I’m currently downloading from the default one at 30kbps (tried googling)
    Or is it the same and apt takes care of architecture?

    Is xorg slow for me only or because mali driver isn’t that good? tried playing a video with mplayer and it was horribly slow

    thanks

    Reply

    • Overclocking only working with android. The clock config in the current kernel is too bad.

      I don’t know other repo.
      Quite fast for me.
      deb http://ports.ubuntu.com quantal main universe restricted multiverse

      Xorg for classic windowing is quite ok on mine but using mplayer is not good.
      Mali is closed source… nearly null support from allwinner, i think it’s the only reason to all this.

      But the mele makes a great headless platform !

      Reply

      • “Overclocking only working with android. The clock config in the current kernel is too bad.”

        Wasn’t the android kernel added to mainline?

        Reply

        • As far as i played with it, debugging and getting the sata stuff work on the mele v1.7 at the time of the 3.0.42, more than 6 months ago, i think there’s some private/closed kernel clock tuning in the stock android… at least in the mele one. It will not be the first time.

          For example, the 3.0.57+ from sunxi is failing to load sata properly. It’s a choice to move forward and they are 100% right to do it.
          I have made mine to upgrade the mali driver in the 3.0.42 which is a fully working kernel for the mele.
          It’s the kernel you can find on this page and i’m encouraging everyone who have the skills and time to work on the 3.0.57 github tree.

          I have not much time anymore to debug, invest and play more on any other a10 platform. I simply bought meles/a10 to make a cluster, i got it fully working.
          The lack of pertinent material around the community made me do this website.

          Now, i’m not sure the guys from Allwinner or Mali or Mele worth it.
          They need to deserve their community.

          Reply

      • echo ‘1152000’ > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq

        That’ll get your freq up to 1.15GHz; up to 1.2GHz is possible without OC and if you are not using any HDD’s.

        Cheers

        Reply

          • No he gets 100fps with egl2gears.
            You can try to follow the specs.

            I did this:

            mv /usr/lib/arm-linux-gnueabihf/mesa-egl /usr/lib/arm-linux-gnueabihf/.mesa-egl
            cd /tmp/
            wget http://dl.linux-sunxi.org/mali/lib-r3p1-01rel0.tar.gz
            mv /lib/libEGL* /lib/libGLE* /lib/libMali.so /lib/libUMP.so ./
            cd /
            tar xvzf /tmp/lib-r3p1-01rel0.tar.gz

            restart xorg.
            2 fps more on glxgears…
            didn’t try egl2gears. I think it’s the trick here.

          • Tried your frequency on my bit perfect mpd player.
            It works for 10 minutes and makes my X act foolish and explode.

            I will stick with the poor 1GHz.
            I don’t want my audio hardware to burn 😉

  • trying to remove plymouth is givig dependency errors (adduser, initscripts, upstart)
    I have a 512MB Mele and only shows 308MB, if the 1GB version is showing 960MB shouldn’t I have 448MB(64 for mali+ramdisk) ?

    Reply

  • Hi, guillaume.
    I want to run headless NAND image on my Mele A2000G 1Gb. I’ve installed
    http://guillaumeplayground.net/mele-a2000-headless-debian-wheezy-armhf-with-nand-install-v1
    to NAND and it’s working ok, but only 512Mb of RAM available. What update i should install to have 1Gb RAM and Ethernet/SATA working?
    Cause there are my results:
    linux-armhf-headless-1G-3.0.42-3.tar.gz – Ethernet is not working , only WiFi. Same issue with IRQ of wemac.
    linux-armhf-gfx-1G-3.0.42-5.tar.gz – Same problem with Ethernet. Even lights aren’t lit at RJ45 port, Only 836Mb of RAM available.

    So, simplest way will be just enable 1Gb in unpatched image. Can you say how to do this? Is it possible to make same with NAND as with SD:
    https://groups.google.com/forum/#!msg/cubieboard/3XT_IdTJ2n0/WcN0rBMS4DEJ
    ?

    Reply

  • New update for GFX on this page.
    Includes some missing modules and the latest sunxi mali.

    Enjoy!

    Reply

  • First of all, thank you for the great contribution you are doing, I feel wonderful. I have a question, and I’m desperate, I have about three weeks trying, even I have installed the latest version of the kernel you have posted on April 4 for a project I’m doing at my university in Spain. I need to use the cubieboard otg port as host, to detect a pl2303 usb to serial device. When I connect to a standard USB port it detects it with no problems detected as ttyUSB0 port for example, but when I connect it to the OTG port appears to me as if powered but no signs of life or mounts. I tried everything possible to use the OTG port as another USB. I’m using to connect to the correct OTG cable and have tried a thousand things but no way, even reinstalled the system several times. When I do a dmesg or lsusb did not show anything. I’m out of ideas, really. Thanks in advance and again, great contribution.

    Reply

    • First, thank you 😉

      I have no otg port on my mele… and no cubieboard.

      The OTG port in classic mode should act as host by default ?
      What’s the output of ‘lsusb -t’ ?
      Perhaps need to mess with the FEL key?
      Did you analyse the script.bin ?
      http://linux-sunxi.org/Fex_Guide#USB_control_flags
      You should be able to force the usb as host only.

      Reply

      • Thanks to you, I’m very gratefully,

        here is the info

        before plug the device:

        root@mele:/# lsusb -t
        /: Bus 04.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
        /: Bus 03.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M
        /: Bus 02.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
        /: Bus 01.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M

        plugin the devide on a normal usb port:

        root@mele:/# lsusb -t
        /: Bus 04.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
        |__ Port 1: Dev 3, If 0, Class=vend., Driver=pl2303, 12M
        /: Bus 03.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M
        /: Bus 02.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
        /: Bus 01.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M

        plugin the device on otg port

        root@mele:/# lsusb -t
        /: Bus 04.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
        /: Bus 03.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M
        /: Bus 02.Port 1: Dev 1, Class=root_hub, Driver=sw-ohci/1p, 12M
        /: Bus 01.Port 1: Dev 1, Class=root_hub, Driver=sw-ehci/1p, 480M

        I can’t find script.bin file, how can I modify it?

        Thanks in advance!

        Reply

        • Ok, i don’t know if your otg port is disable or not.

          To edit your script.bin:


          # mount /dev/mmcblk0p1 /mnt
          # /usr/local/sunxi-tools/bin2fex /mnt/script.bin /mnt/script.orig.fex
          # cp /mnt/script.orig.fex /mnt/script.test.fex

          # vim /mnt/script.test.fex
          << edit # /usr/local/sunxi-tools/fex2bin /mnt/script.test.fex /mnt/script.bin # sync # umount /mnt # reboot

          Reply

          • Thanks again for your help.

            I try several modifications but not works, this is my original script.bin

            [usbc0]
            usb_used = 1
            usb_port_type = 0
            usb_detect_type = 0
            usb_id_gpio =
            usb_det_vbus_gpio =
            usb_drv_vbus_gpio = port:PB09
            usb_host_init_state = 0

            [usbc1]
            usb_used = 1
            usb_port_type = 1
            usb_detect_type = 0
            usb_id_gpio =
            usb_det_vbus_gpio =
            usb_drv_vbus_gpio = port:PH06
            usb_host_init_state = 1

            [usbc2]
            usb_used = 1
            usb_port_type = 1
            usb_detect_type = 0
            usb_id_gpio =
            usb_det_vbus_gpio =
            usb_drv_vbus_gpio = port:PH03
            usb_host_init_state = 0

            [usb_feature]
            vendor_id = 6353
            mass_storage_id = 1
            adb_id = 2
            manufacturer_name = “USB Developer”
            product_name = “Android”
            serial_number = “20080411”

            I change usb_hots_init_state to 1 in usbc2, but still not works and other options, do you have any idea?

            Thanks in advance! I’m ver gratefully!

        • No many other options from my side.
          You can try on cubie’s forum… it’s his board and i don’t have one 😉

          Reply

    • Hola bbellido, este foro es uno de los mas vivos y actulizados sobre avances en la mele, pero me gustaria compartir info y discutir detalles en castellano, como podria hacer para contactarte por mail o chat? soy de la universidad de Cordoba Argentina.

      Reply

  • Hello Again Guillaume,

    how can I do to start automatically X windows and execute an aplication in xwindows? like init applications in windows. There is no .xsessions, what can I do?

    Thanks in advance

    Reply

  • I’m very gratefully, it works great! I do not want to abuse your kindness, but when you run the script in xsession not accept keystrokes, do I need to do something else?

    An example of .xsession

    #!/bin/bash

    xterm

    I need activate something more? Sorry, I know it’s offtopic …

    Best regards

    Reply

  • Hello,

    I have the same problem as Dwayne, I use a cubieboard and only detect 512 Mb, how can I use the 1 GB Memory?

    I have applied UPDATE 5 – Apr 4 2013 kernel update.

    However, great job!!

    Reply

  • There is no solution for the mali problem on cubieboard? I have the latest kernel and tried all: kernels, parameters, etc and nothing work, when I try the modprobe mali, cubie crashes 🙁

    Reply

  • hello, would need to know if you can make the u-boot of the mele display something on screen at boot time?? , at least one flashing bar while booting, as it is very uncomfortable wait for the start without displaying anything on the screen until its all loaded. The ideal would be to see the kernel-verbose, but if you can it at least to indicate that something is booting.

    Reply

  • Hi! After i make linux-armhf-gfx-1G-3.0.42-5.tar.bz2, i’ll get this error on ssh: ssh_exchange_identification: read: Connection reset by peer,
    Any idea what is problem?

    Reply

    • Probably, same specs as the cubieboard?
      I’ve built many different kernels here and in other posts.
      You should try one.

      Reply

  • It’s any chance for a driver for Realtek Semiconductor Corp. RTL8188EUS 802.11n Wireless Network Adapter? Becouse I can’t compile myself.

    Reply

  • Hey Guillaume,
    I found another kernel over at romanrm.net. he is using a newer kernel source. (3.4.79) Any chance, that you will build this newer kernel version too? His images are not working with nand-installs and my kernel-compiling knowledge does not reach that far, that i could compile his kernel to boot from nand. i also would be happy if you could give me a hint, where to start reading about the necessary edits for nand boot. i allready followed the sunxi-linux tutorials on building the bootloader… thank you for this great project, have a nice day

    Reply

  • When updating Debian to the latest testing I get this error:

    °°°°°°°°°°
    Preparing to unpack …/archives/udev_208-8_armhf.deb …
    Since release 198, udev requires support for the following features in
    the running kernel:

    – inotify(2) (CONFIG_INOTIFY_USER)
    – signalfd(2) (CONFIG_SIGNALFD)
    – accept4(2)
    – open_by_handle_at(2) (CONFIG_FHANDLE)
    – timerfd_create(2) (CONFIG_TIMERFD)
    – epoll_create(2) (CONFIG_EPOLL)

    Please upgrade your kernel before or while upgrading udev.

    AT YOUR OWN RISK, you can force the installation of this version of udev
    WHICH DOES NOT WORK WITH YOUR RUNNING KERNEL AND WILL BREAK YOUR SYSTEM
    AT THE NEXT REBOOT by creating the /etc/udev/kernel-upgrade file.
    There is always a safer way to upgrade, do not try this unless you
    understand what you are doing!
    °°°°°°°°°°

    What can I do to solve this without breaking the install?

    Reply
  • Pingback: Cross-compiling « armoid

Leave a Reply

Your email address will not be published. Required fields are marked *