Build kernel from source and boot to Ubuntu using L4T (Linux for Tegra) rootfs

ericvh · Oct 6, 2017

yahoo2016 said:
My SATVs do not work for U3 SD cards. Have you tried U1 SD card or SD cards in USB adapter?

I tried a U1 today with the same problem, will try to find vanilla SD card today to use it with.

GuestK0099 · Nov 4, 2017

had a look into 4.4 seems like some bindings have changed should be fairly possible though,

chtulu · Nov 6, 2017

opendata said:
had a look into 4.4 seems like some bindings have changed should be fairly possible though,

What do you mean some bindings?

I tried everything and it seems impossible.

GuestK0099 · Nov 6, 2017

chtulu said:
What do you mean some bindings?

I tried everything and it seems impossible.

use kernel-tx1 package from https://developer.nvidia.com/embedded/dlc/l4t-sources-28-1 see hardware dir also, it would probably be easier to use the mainline kernel

chtulu · Nov 6, 2017

Done that, and made a missing dtb file for kernel 4.4, it does not go past the nvidia logo.

Without new dtb file, it just reboots.

sunxishan · Nov 7, 2017

JetPack-L4T-3.1-linux-x64.run is with R28.1, it may have some issues here.
You can try to download JetPack 3.0. I tested it and everything works just fine.

guertler said:
Dear Android Hackers,

Short version: L4T in running on my SATV 2015 but CUDA seems not to work correctly.

Long Version: My device is a Shield TV 2015 device and I basically followed the instruction in the first post:

Firmware upgrade after Shield activation to the most recent version

Bootloader unlocking using

Code:

fastboot oem unlock

Flashing "mmcblk1p1.img" obtained in the first post (yahoo2106s Kernel) using

Code:

fastboot flash boot mmcblk1p1.img

Preparing a sdcard according to https://xdaforums.com/showpost.php?p=69202451&postcount=421 and using Tegra210_Linux_R24.2.1_aarch64.tbz2 and Tegra_Linux_Sample-Root-Filesystem_R24.2.1_aarch64.tbz2 and an Ubuntu 14.04 Qemu container

Result was that the SATV boots Linux now! :highfive::good: !! Thanks very much to yahoo2016 and all other contributors to this recipe.

My major interest however is running CUDA on this device. Therefore I

downloaded JetPack-L4T-3.1-linux-x64.run, and run it within an Ubuntu 14.04 Qemu container. There I downloaded cuda-repo-l4t-8-0-local_8.0.84-1_arm64.deb and the cuda-l4t.sh script

used the script on my SATV to install the cuda debian package. It also installed some other dependencies as well.

extended the PATH and LD_LIBRARY_PATH environment variables

compiled (succuessfully) the shipped cuda examples in /usr/local/cuda/samples using make

However the execution of any sample failed: running

Code:

/usr/local/cuda/samples/5_Simulations/nbody/nbody -benchmark -numbodies=65536

only results in the error message

Code:

Error: only 0 Devices available, 1 requested. Exiting.

Strangely also

Code:

lshw -short

does not show the existance of any graphics card hardware:

Code:

H/W path Device Class Description ================================================= system foster_e /0 bus Motherboard /0/0 processor cpu /0/1 processor cpu /0/2 processor cpu /0/3 processor cpu /0/4 memory 2973MiB System memory /1 usb2 bus Nvidia xHCI Host Controller /1/3 generic USB 10/100/1000 LAN /2 usb1 bus Nvidia xHCI Host Controller /3 wlan0 network Wireless interface /4 dummy0 network Ethernet interface /5 enx00044b482f33 network Ethernet interface

On the other hand,

Code:

glxinfo | grep NVIDIA

shows

Code:

server glx vendor string: NVIDIA Corporation client glx vendor string: NVIDIA Corporation OpenGL vendor string: NVIDIA Corporation OpenGL renderer string: NVIDIA Tegra X1 (nvgpu)/integrated OpenGL core profile version string: 4.5.0 NVIDIA 24.2.1 OpenGL core profile shading language version string: 4.50 NVIDIA OpenGL version string: 4.5.0 NVIDIA 24.2.1 OpenGL shading language version string: 4.50 NVIDIA OpenGL ES profile version string: OpenGL ES 3.2 NVIDIA 24.2.1

and glxgears runs with

Code:

13362 frames in 5.0 seconds = 2672.306 FPS

Does anyone have an idea what could be wrong here? I was thinking that maybe

could it be the wrong kernel: Is gpgpu processing not enables in this kernel? I also tested some other kernel images for the SATV 2015 which I found in this forum and on http://www.tatsch.it/linux-4-tegra-with-cuda-on-nvidia-shield-tv/ using

Code:

fastboot boot <kernel>.img

. None of them was able to boot my SATV

Is cuda 8.0 too new and I should use older version (7.0)?

Is this some dtp problem? I did not yet fully understand what this dtp files are for

GuestK0099 · Nov 8, 2017

chtulu said:
Done that, and made a missing dtb file for kernel 4.4, it does not go past the nvidia logo.

Without new dtb file, it just reboots.

You will need to add the shields DTSes (not dtb) to the hardware/something/something dir and diff the Jetson dtses with 3.10 for what you need to change @yahoo2016

chtulu · Nov 8, 2017

opendata said:
You will need to add the shields DTSes (not dtb) to the hardware/something/something dir and diff the Jetson dtses with 3.10 for what you need to change @yahoo2016

Well, I'm not that dumb, I know that, I spent hours, not only to add the missing dts and dtsi, but also to verify them against the existing similar one. Unfortunately with the new dtb file it does not go past the Nvidia logo.

GuestK0099 · Nov 8, 2017

chtulu said:
Well, I'm not that dumb, I know that, I spent hours, not only to add the missing dts and dtsi, but also to verify them against the existing similar one. Unfortunately with the new dtb file it does not go past the Nvidia logo.

I would see if there is any earlyprintk output

chtulu · Nov 9, 2017

opendata said:
I would see if there is any earlyprintk output

Tried earlyprintk, nothing. I do not have a console so I can not see the reason why with the new dtb, the shield does not go past the Nvidia logo. I need someone with shield tv 2017 and console to try the new dtb.

sunxishan · Nov 20, 2017

yahoo2016 said:
mmcblk0p33.img can be download from here:

Please try it use "sudo fastboot boot mmcblk0p33.img"

hi yahoo2016,

When I use this boot image mmcblk0p33.img for new Shield TV 2017, it can boot up without problems but it seems some problem with some internal driver?
when I hooked up a usb camera I cannot see /dev/video0 device.
If I switch back to sda1.img and use roofs on my USB drive everything works just fine. I think it should not be my roofs problem.

when boot from sda1.img,

Code:

ubuntu@tegra-ubuntu:~$ sudo dmesg | grep video
[    7.145279] uvcvideo: Found UVC 1.00 device HD Pro Webcam C920 (046d:082d)

while boot with mmcblk0p33.img,

Code:

ubuntu@tegra-ubuntu:~$ sudo dmesg | grep video
[    0.336250] Linux video capture interface: v2.00

I think here is the problem.

If you can share some hint how to build this boot image maybe i can debug to see what is the issue.

Thanks,

Xishan

crnkoj · Dec 3, 2017

hey guys, been longer away, but still have one of my shields 2015 running the L4T on it (not entirely sure which version though anymore), lsb_release says 16.0.4.1 LTS Xenial.
Now I was wondering if I can update it without breaking it and which packages need to be held back? Thanks!

yahoo2016 · Dec 4, 2017

crnkoj said:
hey guys, been longer away, but still have one of my shields 2015 running the L4T on it (not entirely sure which version though anymore), lsb_release says 16.0.4.1 LTS Xenial.
Now I was wondering if I can update it without breaking it and which packages need to be held back? Thanks!

L4T R24.2.1 is still the last known working version for SATV.

I have been trying L4T R28 on Jetson TX1/TX2, there are many issues.

There is no DTS for SATV (foster) in L4T R28. It appears Nvidia dropped L4T support for SATV since release of TX2.

crnkoj · Dec 4, 2017

yahoo2016 said:
L4T R24.2.1 is still the last known working version for SATV.

I have been trying L4T R28 on Jetson TX1/TX2, there are many issues.

There is no DTS for SATV (foster) in L4T R28. It appears Nvidia dropped L4T support for SATV since release of TX2.

Thanks. But what about the Ubuntu updates through apt-get, can I do those? BTW are you updating the latest usable kernel with security updates?
edit> one more thing, did we give up on chainloading o-boot after the android bootloader? This guy got it working for the old OMAP devices, perhaps useable ideas?
https://xdaforums.com/galaxy-nexus/development/bootloader-boot-multi-boot-support-t2201146
https://xdaforums.com/galaxy-s2/general/uboot-bootloader-true-multiboot-t1680898

yahoo2016 · Dec 6, 2017

crnkoj said:
Thanks. But what about the Ubuntu updates through apt-get, can I do those? BTW are you updating the latest usable kernel with security updates?
edit> one more thing, did we give up on chainloading o-boot after the android bootloader? This guy got it working for the old OMAP devices, perhaps useable ideas?
https://xdaforums.com/galaxy-nexus/development/bootloader-boot-multi-boot-support-t2201146
https://xdaforums.com/galaxy-s2/general/uboot-bootloader-true-multiboot-t1680898

I do not see issues for "sudo apt-get update".
I have not touched my SATVs for long time. I have been busy with TX1s/TX2s at work.

crnkoj · Dec 6, 2017

yahoo2016 said:
I do not see issues for "sudo apt-get update".
I have not touched my SATVs for long time. I have been busy with TX1s/TX2s at work.

Thanks mate.
Speaking of tx1/2, I saw the tx1 costs 239e now, whole the tx2 costs 600e. Is everything working fine with those? Is it possible to compile kernels natively on those, eventually use a different distronic than Ubuntu with working nvidia drivers (eg like for desktop cards in PCs for instance with gentoo)?

yahoo2016 · Dec 8, 2017

crnkoj said:
Thanks mate.
Speaking of tx1/2, I saw the tx1 costs 239e now, whole the tx2 costs 600e. Is everything working fine with those? Is it possible to compile kernels natively on those, eventually use a different distronic than Ubuntu with working nvidia drivers (eg like for desktop cards in PCs for instance with gentoo)?

Unless more GPU cores are needed, TX1s are better valued. TX1s are compatible with earlier versions of L4Ts, TX2s only support L4T 27 and later.
There are significant differences between kernel of L4T 24.2.1 and kernels of L4T 27 and later.
L4T 24.2.1 has been stable and everything seems working (the same for SATV) as far as I know.
There have been many issues reported for TX2s on Nvidia Devtalk.
I have been natively compiling L4T kernels for both TX1s and TX2s. 32 bit compilers are not needed for later L4Ts.
We are more hardware oriented to use TX1s/TX2s for sensor (e.g., MIPI cameras, IMUs) related work and have not tried to use OSs other than L4T. The vendors only supply drivers for L4T.

Mogster2K · Jan 9, 2018

Will the kernels in the first post work for L4T 24.2.2? Also, which file(s) need to be changed to select the rootfs location?

deerish · Jan 21, 2018

chtulu said:
Tried earlyprintk, nothing. I do not have a console so I can not see the reason why with the new dtb, the shield does not go past the Nvidia logo. I need someone with shield tv 2017 and console to try the new dtb.

Please provide your work thus far. The UART is on the debug connector next to the fan header.

chtulu · Jan 22, 2018

Please provide your work thus far. The UART is on the debug connector next to the fan header.

Hi, I have already lost one shield tv, trying to get the last kernel working, so I am not going to use the UART, besides the shield is running 24/7 with cuda code, so no chance.

I have made the new dtb for shield tv 2017 and not the one this thread is about, but as I said the shield is stuck at the logo when using it, so I gave up. If someone want to have a look to the entire source code dtb, dts etc I can upload somewhere.

Build kernel from source and boot to Ubuntu using L4T (Linux for Tegra) rootfs

Member

GuestK0099

Guest

Member

GuestK0099

Guest

Member

New member

GuestK0099

Guest

Member

GuestK0099

Guest

Member

New member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Member

Member

Similar threads

Top Liked Posts