Feedback and Queries regarding Armbian 23.11 Bookworm CLI Kernel 6.7 on Orange Pi 5

[Marco Schirrmeister]

I know now why sshd is not running. The directory /run/sshd is missing. dpkg-reconfigure openssh-server or service ssh restart from armbian-firstrun delete the directory if it is there.

Just restarting sshd via the console recreates it (as defined in its service).

 

sshd -t
Missing privilege separation directory: /run/sshd

 

It can easily reproduced by enabling the armbian-firstrun service and a reboot. It runs through its things on the next boot and the directory is gone.

I assume it has to do with the way the sshd restart is executed. Something where it is executed in a script, that in turn is executed via a systemd service during a boot. If you restart it via the console shell where you logged in, it works just fine. 

 

Edited January 30 by Marco Schirrmeister

[obiwantoby]

I have to say the latest ./compile.sh with 6.8.0-rc1-edge-rockchip-rk3588 seems to be operating great. 

 

For some reason the benchmarks have increased dramatically from 5.10-legacy. Ping times are a big lower overall too on average - unsure if the big.LITTLE scheduling improvements are contributing - but in ARM those should have been fairly mature, not like the Adler Lake e core changes. Either way I am welcoming the change.

 

https://pibenchmarks.com/benchmark/78720/

 

19,000 --> 32,000

 

     Category                  Test                      Result     

HDParm                    Disk Read                 376.81 MB/sec            

HDParm                    Cached Disk Read          386.25 MB/sec            

DD                        Disk Write                270 MB/s                 

FIO                       4k random read            86050 IOPS (344201 KB/s)

FIO                       4k random write           46972 IOPS (187889 KB/s)

IOZone                    4k read                   172508 KB/s              

IOZone                    4k write                  131029 KB/s              

IOZone                    4k random read            69041 KB/s               

IOZone                    4k random write           159514 KB/s              

 

                          Score: 32652                                       

[Marco Schirrmeister]

Final comment for the sshd start problem on some systems during the first boot. That should probably discussed in its own thread though.

It really seems to come down to the start time of the armbian-firstrun service. I guess it starts too early.

 

Either of the following changes in the armbian-firstrun.service file will have it started later and avoids the issue. 

 

[Service]
Type=idle

 

or

 

[Unit]
After=multi-user.target

 

Edited February 2 by Marco Schirrmeister
Formatting

[David Pottage]

I have noticed that the 6.8 kernel on my OrangePi5 updates more than once per day (via the Debian package apt update mechanism)

 

1. The version number of the kernel still reports as RC1, even though kernel.org released RC2 more than a week ago, and RC3 a few days back. Does the Armbain build system need a PR to a config file to update the upstream kernel that it pulls and builds.

root@orangepi5:~# uname -a
Linux orangepi5 6.8.0-rc1-edge-rockchip-rk3588 #2 SMP PREEMPT Sun Jan 21 22:11:32 UTC 2024 aarch64 GNU/Linux

2. Are all the kernel releases actually different? There appears to be a huge amount of churn.

 

[Marco Schirrmeister]

The edge or mainline kernel version is defined in config/sources/mainline-kernel.conf.sh. So yes, an update/PR to that file is needed, I think.

If you want to change to your own newer version, you can define it for the compile.sh script.

 

builder@dumpster /m/t/t/n/build (main)# cat userpatches/lib.config
KERNELBRANCH="tag:v6.8-rc2"

 

[Werner]

Thought about bumping to rc3 yesterday but figured it wasn't worth it since there were no relevant changes.

[Gullik]

Does that mean a new kernel is also in the update/upgrade procedure, or is it just the prebuilt edge dists?

 

Gullik

[Marco Schirrmeister]

If you installed an image from the nightly builds, then the apt source list should have the beta repo in it. Which means you will see kernel and other updates relative often. 

 

root@orangepi5-plus ~# cat /etc/apt/sources.list.d/armbian.list
deb [signed-by=/usr/share/keyrings/armbian.gpg] http://beta.armbian.com trixie main trixie-utils trixie-desktop

 

[David Pottage]

11 hours ago, Marco Schirrmeister said:

If you installed an image from the nightly builds, then the apt source list should have the beta repo in it. Which means you will see kernel and other updates relative often. 

I investigated, and found that many nightly builds are identical other than the version number.

 

Using a recent range of version numbers still cached in my board's filling system:  468 469 473 474 475 477 480 484 490 491 503 516 517 520 525 527

I extracted all the package control files:

for ver in 468 469 473 474 475 477 480 484 490 491 503 516 517 520 525 527 ; do mkdir -p /tmp/extract/24.2.0-trunk.${ver}_arm64 ; dpkg-deb -e /var/cache/apt/archives/linux-image-edge-rockchip-rk3588_24.2.0-trunk.${ver}_arm64.deb  /tmp/extract/24.2.0-trunk.${ver}_arm64 ; done

See also: https://forums.debian.net/viewtopic.php?t=60431

 

Those version numbers cover a date range from 27th Jan to 5th Feb

 

I then got the checksum of each md5sums file to see when it changed:

 

root@orangepi5:/tmp# sha1sum /tmp/extract/*/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.468_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.469_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.473_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.474_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.475_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.477_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.480_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.484_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.490_arm64/md5sums
b095c3719b4682c6f7f2d9809effeb77bb5f9a20  /tmp/extract/24.2.0-trunk.491_arm64/md5sums
64ac8a009665064bbc877777f9f1a0e1d847fcf8  /tmp/extract/24.2.0-trunk.503_arm64/md5sums
64ac8a009665064bbc877777f9f1a0e1d847fcf8  /tmp/extract/24.2.0-trunk.516_arm64/md5sums
64ac8a009665064bbc877777f9f1a0e1d847fcf8  /tmp/extract/24.2.0-trunk.517_arm64/md5sums
64ac8a009665064bbc877777f9f1a0e1d847fcf8  /tmp/extract/24.2.0-trunk.520_arm64/md5sums
64ac8a009665064bbc877777f9f1a0e1d847fcf8  /tmp/extract/24.2.0-trunk.525_arm64/md5sums
64ac8a009665064bbc877777f9f1a0e1d847fcf8  /tmp/extract/24.2.0-trunk.527_arm64/md5sums

 

So from this we can see that the contents of the package only changed once over that time period. Other than that the only difference between package versions is the version number itself in the DEBIAN/control file.

 

Given this, would it make sense to add a step to the build and publish scripts to skip publishing a new version of a package if it's contents are identical to the previous version? I am working on Debian packaging scripts for work anyway so it would not be hard.

 

[Gullik]

Is there any reason to keep kernel edge at rc1 rather than rc2 or rc3?

 

I for one would like to see the latest kernel improvements for rk3588.

 

I guess building armbian ON armbian is not supported, ( due to compiler issue??) or am I wrong??

else I could try this out myself.

 

Gullik

[Gunjan Gupta]

3 hours ago, Gullik said:

I guess building armbian ON armbian is not supported, ( due to compiler issue??) or am I wrong??

Depends on boards and vendors. Raspberry pi and allwinner works, Khadas vim1s and vim4 doesn't. Not sure about rockchip boards, I don't have any so haven't tried. Some vendors uses closed source toolchains with binaries available only on x86 platform for building bootloaders or kernels. Just give building a try, if there is a restriction like that, it will let you know.

 

[Marco Schirrmeister]

Building on an Orange Pi 5 works just fine. I build image for the OPi5+ and Rock5b on my OPi5.

 

@Gullik, what improvements do you expect to see in rc2 or rc3? Like Werner wrote the other day, he has not changed it to rc2, since there were no relevant changes.

Same most likely goes for rc3. There are no commits for rk3588 or at least none where you would say, I need this right now. Even if there are changes, many are board specific. 

 

[Gullik]

And a comment to Marco, it not so easy to understand all technicalities in the workflow so bear with my ignorance.

I have built linux since V 1.0, but not in the last 20 years, so a lot has changed, and dists are assembled from many

sources and groups. However, I feel mych better at home now, read my posting, Regards, Gullik.

 

 

Beeing really adventurous I have attempted to build armbian ON armbian 6.8-rc1 noble.

This is done on an Orange pi 5 8Gb, booted from an SD card.

The machine has a 256 Gb nvme memory and a 8821 wifi adapter

In the build documentation it is stated that I need:

 

"A x86/x64 machine running any OS; at least 4G RAM, SSD, quad core (recommended)"
"The officially supported compilation environment is Ubuntu Jammy 22.04.x amd64 only!"

 

I understand that having something supported rather than working are quite different,
and that Armbian have higher standards than me as a user, but I did not get discouraged.

These are my modifications to various components to get armbian to compile.

So, here goes:

 

1) git clone --depth=1 --branch=main https://github.com/armbian/build

2) add noble to supported build environments

vi lib/functions/host/host-release.sh

        # Disable host OS check at your own risk. Any issues reported with unsupported releases will be closed without discussion
--        if [[ -z $HOSTRELEASE || "bookworm trixie sid jammy kinetic lunar vanessa vera victoria mantic" != *"$HOSTRELEASE"* ]]; then
++        if [[ -z $HOSTRELEASE || "bookworm trixie sid jammy kinetic lunar vanessa vera victoria mantic noble" != *"$HOSTRELEASE"* ]]; then
                if [[ $NO_HOST_RELEASE_CHECK == yes ]]; then
                        display_alert "You are running on an unsupported system" "${HOSTRELEASE:-(unknown)}" "wrn"
                        display_alert "Do not report any errors, warnings or other issues encountered beyond this point" "" "wrn"
                else
                        exit_with_error "Unsupported build system: '${HOSTRELEASE:-(unknown)}'"
                fi
        fi
}

3) I ignored warnings that I did not have 10 Gb available in /output directory and pressed return

4) vi config/sources/mainline-kernel.conf.sh
    change 6.8-rc1 to 6.8-rc3 which was the desired kernel
5) change requirements from python2 to python3    // can cause unforeseen problems but seems to work

I had to accept "shallow-tree" for kernel due to lack of space.......

and the result after booting:

Welcome to Armbian-unofficial 24.2.0-trunk Bookworm with bleeding edge Linux 6.8.0-rc3-edge-rockchip-rk3588

No end-user support: built from trunk

System load:   14%               Up time:       9 min    
Memory usage:  4% of 7.51G      IP:           192.168.1.89
CPU temp:      32°C               Usage of /:    1% of 233G       
RX today:      26.2 KiB      

These 3 changes enabled me to run the ./compile.sh script. I selected a simple server only
config in the menus, since I was mostly interested in testing USB ports.

Unfortunately the USB fixes that are documented for the rk3588 did not result in
the addition of the vertical usb port, so something is wrong, but I can build my own kernel.

Interesting to see is that all 8 cpus are running at >=90%, so this $100 machine replaces
a big Proliant and performs much better. Having learned programing on a PDP11/40 with
about 8 users, this is mindboggling. Also replacing the "House server" from a 1U server
to an Orange PI 3 LTS paid for the board in reduced electricity bill in a few months,
and runs on a used 12 V car battery.

 

My sincere compliments to a solid source tree, and a highly automated build environment,
these are probably the main reason for success on this foolish attempt!!

Next step is to install "noble" to the NVME, to get rid of the output size warning, and
build a new desktop with upgraded kernel.

 

Best regards, and an applause for the Armbian Crew!

 

Gullik

[Gullik]

I can see that Sebastian Reichel has done a lot of work to usb in the rk3588 kernel.

This is as I understand to the "current" or latest kernel.

 

Is the Armbian kernel (edge) built from this tree, or is review done on a separate

tree?

 

Regards,

Gullik

[Werner]

edge builds from mainline with these patches on top: https://github.com/armbian/build/tree/main/patch/kernel/rockchip-rk3588-edge

Feel free to port some of Sebastians stuff which did not make it into 6.8.

[Gullik]

Thanks Werner for the explanation.

 

I guess then that Sebastians patches will be included with the next rc.

 

Gullik

[Werner]

Those which were sent upstream and accepted certainly will. It all depends on how long you want to wait seeing it running in Armbian

 

[sugatam]

Isn't the reason you created the "collabora" branch? Its unfortunately stuck at 6.7.0, but I was happily using it as it was getting all of Sebastian's (and team collabora's) updates before they were accepted upstream.