VOXL2 kernel version

jcai

So I was reading through the kernel build guide, and I saw an example that enabled smsc95xx by modifying the m005x.cfg file (CONFIG_USB_NET_SMSC95XX=y).

From what I can tell by reading through the IPVLAN docs, it's much the same steps to enable it (CONFIG_IPVLAN=y or CONFIG_IPVLAN=m). So it looks like it should be safe? Is this still something you would advise against? Also, would there be any issues with running all of the build stuff on MacOS?

Alex Kushleyev

@jcai ,

If you think you might want to try building the kernel and enabling a certain module, you could try it without a risk of bricking your VOXL2. There is a way to test a kernel without overwriting your existing kernel, so if something goes wrong, you just need to power cycle voxl2 and you are good. In the test mode, the kernel is loaded in temporary memory and is then launched.

https://docs.modalai.com/voxl2-kernel-build-guide/#test

Specifically, use the following command:

adb reboot bootloader
fastboot boot your_new_kernel.img

Do not use fastboot flash commands until you are sure you want to overwrite the current kernel.

Regarding running the build on MacOS, since OSX (assuming x86_64) is not running on the same kernel type, Docker will have to emulate the Linux kernel (not using host kernel) because that is what the kernel build docker image uses. It should still work, but building will be slower. However, if you have Parallels, you can run a Ubuntu VM, which will interact with the CPU directly, bypassing OSX, so no emulation is needed, and your build will be very fast (assuming x86_64 mac). If you have ARM64 mac, then Parallels will probably not let you run x86_64 Ubuntu VM, so your best bet is just to use native ARM64 Docker application with x86_64 translation enabled (and run our x86_64 kernel build image), which may be pretty fast.

I believe we have had users port a driver from a newer kernel by just updating the source code from a newer kernel into our 4.19.125, but there is no guarantee it will work, depending on the scope of changes.

Alex

jcai

@Alex-Kushleyev I'm having some issues with building the kernel. I'm running the latest Ubuntu in VirtualBox (on Windows) for this. In the container, the sync and patch scripts run without issue but when building as root, I get the following error:

If I run build without sudo, I get access permission issues:

Alex Kushleyev

@jcai , i will try to do the build on native linux machine just in case. Can you confirm you did not make any changes to the kernel source or build scripts before building?

Alex

jcai

@Alex-Kushleyev Yeah, no changes

Alex Kushleyev

@jcai , I was able to successfully build the m0054 kernel by following the exact instructions from here : https://docs.modalai.com/voxl2-kernel-build-guide/

The first time, i did get an error:

Cloning into bare repository '/home/user/build_mount/lu.um.1.2.1/apps_proc/downloads/git2/github.com.dex4er.fakechroot.git'...
fatal: unable to access 'https://github.com/dex4er/fakechroot.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated.

but this must have been a temporary web glitch, as it worked by just re-running the build.

When you say that you ran the build as root, how exactly did you do that? Did you run as root first then as non-root? if so, it is possible that root user created directories that are not writable by normal user, in which case you may need to clean and build from scratch.

Alex

jcai

@Alex-Kushleyev I was only able to follow the guide exactly until I got into the container. From there , all the scripts inside were run with sudo. Otherwise, I would get permission issues like below.

Alex Kushleyev

@jcai , as this is a permissions issue, can you please check the following:

inside docker

ls -lah /home/user/build_mount/lu.um.1.2.1/apps_proc

and outside docker

ls -lah ~/qrb5165-kernel-build-docker/workspace/lu.um.1.2.1/apps_proc

In my case, all the folders that were created inside the docker container are owned by my actual username on the host machine (and permissions are drwxr-xr-x for the directory apps_proc and others. This means only the owner can write to it. But inside docker, the ownership is shown as user user, so docker is writing as user and on the host machine it is written on my actual username's behalf.

user@098ef18e8424:~/build_mount/lu.um.1.2.1/apps_proc$ ls -lah
total 80K
drwxr-xr-x  10 user user 4.0K Oct  9 15:45 .
drwxr-xr-x   3 user user 4.0K Oct  9 15:31 ..
drwxr-xr-x   6 user user 4.0K Oct  9 16:31 build-qti-distro-ubuntu-fullstack-perf
drwxr-xr-x   2 user user  20K Aug 23  2022 deb_premirror_packages
drwxr-xr-x   5 user user 4.0K Oct  9 15:37 disregard
drwxr-xr-x  13 user user  28K Oct  9 16:29 downloads
drwxr-xr-x  34 user user 4.0K Oct  9 15:38 poky
drwxr-xr-x   7 user user 4.0K Oct  9 15:37 .repo
lrwxrwxrwx   1 user user   27 Oct  9 15:38 setup-environment -> poky/qti-conf/set_bb_env.sh
drwxr-xr-x  25 user user 4.0K Oct  9 15:38 src
drwxr-xr-x 177 user user 4.0K Oct  9 16:29 sstate-cach

It is possible that in your case something is not working, i am curious what the ownership shows.

Alex

jcai

On the host:

jcai@jcai-VirtualBox:~/qrb5165-kernel-build-docker/workspace/lu.um.1.2.1/apps_proc$ ls -lah
total 48K
drwxr-xr-x  8 root root 4.0K Oct  9 15:42 .
drwxr-xr-x  3 root root 4.0K Oct  8 13:03 ..
drwxrwxr-x  2 1001 1001  20K Aug 23  2022 deb_premirror_packages
drwxr-xr-x  5 root root 4.0K Oct  8 17:27 disregard
drwxr-xr-x  2 root root 4.0K Oct  9 09:49 downloads
drwxr-xr-x 34 root root 4.0K Oct  9 15:25 poky
drwxr-xr-x  7 root root 4.0K Oct  9 15:23 .repo
lrwxrwxrwx  1 root root   27 Oct  8 17:33 setup-environment -> poky/qti-conf/set_bb_env.sh
drwxr-xr-x 25 root root 4.0K Oct  8 17:31 src

In Docker:

user@8de0065c2813:~/build_mount/lu.um.1.2.1/apps_proc$ ls -lah
total 48K
drwxr-xr-x  8 root root 4.0K Oct  9 19:42 .
drwxr-xr-x  3 root root 4.0K Oct  8 17:03 ..
drwxrwxr-x  2 1001 1001  20K Aug 23  2022 deb_premirror_packages
drwxr-xr-x  5 root root 4.0K Oct  8 21:27 disregard
drwxr-xr-x  2 root root 4.0K Oct  9 13:49 downloads
drwxr-xr-x 34 root root 4.0K Oct  9 19:25 poky
drwxr-xr-x  7 root root 4.0K Oct  9 19:23 .repo
lrwxrwxrwx  1 root root   27 Oct  8 21:33 setup-environment -> poky/qti-conf/set_bb_env.sh
drwxr-xr-x 25 root root 4.0K Oct  8 21:31 src

Yeah, ownership looks like an issue here

Alex Kushleyev

@jcai , yes it looks like all the folders have root ownership, and the build script does not like being run as root, however non-root cannot write to those folders.

Please try to run the build from scratch, removing the workspace folder and running the sync, patch, build scripts as non-root. After you sync, please check the ownership of the folders, if root is the owner, the same issue will happen.

If after build and sync, the folders are owned by root, you can try to do the following from your docker container after running sync and patch (which seem to work fine, right?):

sudo chown -R user:user ~/build_mount/

This will make user the owner of those files and folders (recursively)

Alternatively you can assign all permissions to all users for this build folder

sudo chmod -R 777 ~/build_mount/

And then run the actual build as user

Alex

jcai

@Alex-Kushleyev I've managed to get the build working, thanks! I think the problems started because my initial docker image was built with sudo. Removing the workspace folder and starting from scratch got everything set up as user.

I'm now attempting to test the build using:

adb reboot bootloader
fastboot boot your_new_kernel.img

Are there specific drivers I need to download? Once I put the voxl into bootloader, VirtualBox can't detect the device anymore. The same is true on my Windows host. As a result, running fastboot boot new_kernel.img gives me a constant <waiting for any device> message. As reference, here's my device manager during normal and bootloader modes. The "Android" device also comes and goes. It usually exists for a couple seconds before disappearing.

Alex Kushleyev

@jcai , sometimes fastboot requires running at it as root, so please try

sudo fastboot boot your_new_kernel.img

jcai

@Alex-Kushleyev I think this is just an issue with Windows drivers. I ended up switching over to OSX to use adb and fastboot, no issues there. I'm just using a USB drive to get the .img file from the VM to the Mac. Boot tests look good.

One more thing, I'm looking at the kernel build guide and there's a section that says I can enable in tree drivers by editing:

apps_proc/poky/meta-voxl2-bsp/recipes-kernel/linux-msm/files/m005x.cfg

I don't have this file in my system. the closest I found is ../linux-msm/files/configs/m0054. In this directory, there's kona_defconfig and kona-perf_defconfig. What's the difference between these two files? Are these the correct files to be editing to enable, for example, CONFIG_MACVLAN?

Alex Kushleyev

@jcai ,

In my kernel build, the config file exists:

workspace$ find . -name m005x.cfg
./lu.um.1.2.1/apps_proc/build-qti-distro-ubuntu-fullstack-perf/tmp-glibc/work/m0054-oe-linux/linux-msm/4.19-r0/fragments/m005x.cfg
./lu.um.1.2.1/apps_proc/poky/meta-voxl2-bsp/recipes-kernel/linux-msm/files/fragments/m005x.cfg

kona-perf_defconfig is the "performance" kernel, which is what you should be using
kona_defconfig is the debug version of the kernel configuration

I believe you can edit either the m005x.cfg config fragment or kona-perf_defconfig , since linux-msm_4.19.bbappend has the following:

do_configure_prepend() {
   # We merge the config changes with the default config for the board
   # using merge-config.sh kernel tool

   mergeTool=${S}/scripts/kconfig/merge_config.sh
   confDir=${S}/arch/${ARCH}/configs
   defconf=${confDir}/${KERNEL_CONFIG}

   ${mergeTool} -m -O ${confDir} ${defconf} ${CFG_FRAGMENTS}

   # The output will be named .config. We rename it back to ${defconf} because
   # that's what the rest of do_configure expects
   mv ${confDir}/.config ${defconf}
   bbnote "Writing back the merged config: ${confDir}/.config to ${defconf}"
}

which is merging the config fragment with the defconfig.

I hope this helps!

Alex

jcai

@Alex-Kushleyev Yup, editing kona-perf_defconfig looks to have worked, thanks!