Kernel 5.4 Development Party

[memeka]

Continuing the discussion from viewtopic.php?f=95&t=33510&p=273298#p273298

Let's start the 5.4 kernel development party.

I'm pushing my 5.4 branch based on 5.4.0 at: https://github.com/mihailescu2m/linux/t ... dxu4-5.4.y



HC-1 sd-card test:
* write

Code: Select all

64+0 records in
64+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 22.0655 s, 24.3 MB/s

* read

Code: Select all

64+0 records in
64+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 6.36601 s, 84.3 MB/s

HC-1 SSD test:
* write

Code: Select all

64+0 records in
64+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 5.72404 s, 93.8 MB/s

* read

Code: Select all

64+0 records in
64+0 records out
536870912 bytes (537 MB, 512 MiB) copied, 1.39248 s, 386 MB/s

[odroid]

@memeka,
Thank you for making a new topic.
I've created a dedicated sub-forum and moved your post to here to enjoy this dev party with more people and topics.

Can you please write a defconfig file which can run on our generic Ubuntu Mate X11 desktop image?

[memeka]

Can you please write a defconfig file which can run on our generic Ubuntu Mate X11 desktop image?

it's in there aleady, odroidxu4_defconfig (last commit)

[odroid]

Great! We will try it on our latest Ubuntu 18.04.3 Mate image.
https://wiki.odroid.com/odroid-xu4/os_i ... 4/20190929

[joshua.yang]

I'm going to write about how to use this modern kernel.

1. Download our official Ubuntu image from this link
- Mate: https://wiki.odroid.com/odroid-xu4/os_i ... 4/20190929
- Minimal: https://wiki.odroid.com/odroid-xu4/os_i ... 10-minimal

2. Flash that image using Etcher to your SD card / eMMC.

3. Insert that card to your Odroid XU4 for the initial boot sequence.
3.1. It will resize its root file system to fit into your flash memory capacity.
3.2. Then do the package upgrade. Enter apt update && apt full-upgrade -y. If it failed with a message that seems trying to say locking problem, wait for about 5~10 minutes and try again.

4. Make linux-kernel-5422 package not to be upgraded. Enter apt-mark hold linux-odroid-5422.

5. Reboot.

6. Power off and connect your SD card / eMMC to your PC.

Now let's start to build 5.4 kernel for XU4.

1. Setup build environment by referring to this guide: https://wiki.odroid.com/odroid-xu4/soft ... ross-build
1.1 Download the proper toolchain.
1.2 Export environment variables to do with using the toolchain.

2. Mount boot, rootfs partitions of SD card / eMMC to your PC. It should be mounted automatically on Ubuntu.

3. Clone @memeka's 5.4 kernel. Enter git clone https://github.com/mihailescu2m/linux.git --depth 1 -b odroidxu4-5.4.y linux-kernel-odroidxu4-5.4.y.

4. Move to the cloned directory. Enter cd linux-kernel-odroidxu4-5.4.y.

5. Build 5.4 kernel. Cheer for the computer.

Code: Select all

make odroidxu4_defconfig
make -j $(nproc)

6. Copy kernel image and device tree blob to the media card's boot partition. Replace target path to yours.

Code: Select all

sudo cp -f arch/arm/boot/zImage /media/joshua/boot
sudo cp -f arch/arm/boot/dts/exynos5422-odroidxu4.dtb /media/joshua/boot

7. Install modules.

Code: Select all

sudo make -j $(nproc) modules_install ARCH=arm INSTALL_MOD_PATH=/media/joshua/rootfs

8. Do sync and unmount your boot media card and insert it to XU4.

Code: Select all

root@odroid:~# uname -a
Linux odroid 5.4.0+ #2 SMP Wed Nov 27 08:17:23 UTC 2019 armv7l armv7l armv7l GNU/Linux
root@odroid:~# lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 18.04.3 LTS
Release:        18.04
Codename:       bionic

Enjoy
And many thanks to @memeka.

[joshua.yang]

Can you please write a defconfig file which can run on our generic Ubuntu Mate X11 desktop image?

it's in there aleady, odroidxu4_defconfig (last commit)

Hi @memeka,

In my case, with 5.4 kernel it shows nothing but only blinking cursor at the left top.
Any other configurations I have to do? Please check the guides above.

[rooted]

I've already ran into a problem, the ATSC drivers I need aren't available in 5.4.y it seems.

I know I need these:

Code: Select all

CONFIG_DVB_AU8522=m
CONFIG_DVB_AU8522_DTV=m
CONFIG_DVB_AU8522_V4L=m

I guess I'm out of the testing until these make it in unfortunately as I must have ATSC.

[memeka]

Hi @memeka,

In my case, with 5.4 kernel it shows nothing but only blinking cursor at the left top.
Any other configurations I have to do? Please check the guides above.

this is how i do it, on a standard ubuntu system where /media/boot is the boot partition:

Code: Select all

cp arch/arm/config/odroidxu4_defconf .config
make -j7 zImage modules dtbs

sudo cp arch/arm/boot/zImage /media/boot/zImage
sudo cp arch/arm/boot/dts/exynos5422-odroid*.dtb /media/boot
sudo make modules_install
kver=`make kernelrelease`
sudo cp .config /boot/config-${kver}
cd /boot
sudo update-initramfs -c -k ${kver}
sudo mkimage -A arm -O linux -T ramdisk -a 0x0 -e 0x0 -n initrd.img-${kver} -d initrd.img-${kver} uInitrd-${kver}
sudo cp uInitrd-${kver} /media/boot/uInitrd

[rooted]

Do you know a timeframe when the rest of the USB media drivers will make it @memeka?

Surely the device I use isn't depreciated as it's one of the most common ATSC tuners in use.

[mad_ady]

@rooted isn't it just a kernel config issue?

[joshua.yang]

this is how i do it, on a standard ubuntu system where /media/boot is the boot partition:

Code: Select all

cp arch/arm/config/odroidxu4_defconf .config
make -j7 zImage modules dtbs

sudo cp arch/arm/boot/zImage /media/boot/zImage
sudo cp arch/arm/boot/dts/exynos5422-odroid*.dtb /media/boot
sudo make modules_install
kver=`make kernelrelease`
sudo cp .config /boot/config-${kver}
cd /boot
sudo update-initramfs -c -k ${kver}
sudo mkimage -A arm -O linux -T ramdisk -a 0x0 -e 0x0 -n initrd.img-${kver} -d initrd.img-${kver} uInitrd-${kver}
sudo cp uInitrd-${kver} /media/boot/uInitrd

Thanks for your answer.

It seems the only difference is ramdisk thing..

Actually, I think I forgot to say the point. I can see the console screen on the HDMI monitor with Minimal image, but it isn't with Mate image. I can connect to XU4 with both of them using SSH.
The only blinking cursor issue is with the Mate image.

Could you check it again? Thanks in advance.

[rooted]

@rooted isn't it just a kernel config issue?

Unfortunately it isn't, the drivers are missing. They aren't enabled in memeka's configuration so I went into menuconfig to enable them but they simply aren't there.

I even tried to enable them manually and then running 'make oldconfig' but it simply removed them from the config.

[mad_ady]

So they don't show up even if you search for them in menuconfig with /, do they? I haven't heard of drivers being removed from the kernel, except maybe if there was a big change in the api and there was no maintainer left for those bits... How odd...

[memeka]

@rooted: https://cateee.net/lkddb/web-lkddb/DVB_AU8522.html

Code: Select all

Found in Linux kernels: 3.7–3.19, 4.0–4.20, 5.0–5.4, 5.4+HEAD

So it's in there.

Code: Select all

depends on: CONFIG_DVB_CORE && CONFIG_I2C

Make sure CONFIG_DVB_CORE and CONFIG_I2C are enabled, then the option will appear

[rooted]

@rooted: https://cateee.net/lkddb/web-lkddb/DVB_AU8522.html

Code: Select all

Found in Linux kernels: 3.7–3.19, 4.0–4.20, 5.0–5.4, 5.4+HEAD

So it's in there.

Code: Select all

depends on: CONFIG_DVB_CORE && CONFIG_I2C

Make sure CONFIG_DVB_CORE and CONFIG_I2C are enabled, then the option will appear

I will look again, thanks.

[odroid]

Hi @memeka,

In my case, with 5.4 kernel it shows nothing but only blinking cursor at the left top.
Any other configurations I have to do? Please check the guides above.

this is how i do it, on a standard ubuntu system where /media/boot is the boot partition:

Code: Select all

cp arch/arm/config/odroidxu4_defconf .config
make -j7 zImage modules dtbs

sudo cp arch/arm/boot/zImage /media/boot/zImage
sudo cp arch/arm/boot/dts/exynos5422-odroid*.dtb /media/boot
sudo make modules_install
kver=`make kernelrelease`
sudo cp .config /boot/config-${kver}
cd /boot
sudo update-initramfs -c -k ${kver}
sudo mkimage -A arm -O linux -T ramdisk -a 0x0 -e 0x0 -n initrd.img-${kver} -d initrd.img-${kver} uInitrd-${kver}
sudo cp uInitrd-${kver} /media/boot/uInitrd

@memeka,
Do you run Kernel 5.4 on Mate/x11 18.04.3 image?

[memeka]

No, I only tried wayland - Weston and Gnome3

[MastaG]

Kernel running great here
Still have to look into the segfault when going into suspend on the XU3 (which is probably related to the ethernet driver).

For anyone having a hard time with the kernel configuration, I've attached mine which I'm using for Fedora 31.
It's based on Fedora's kernel configuration so it differs a lot from the default odroidxu4_defconfig.

linux-5.4.0.config.zip

(41.17 KiB) Downloaded 208 times

The following stuff is enabled:
- Virtualization (platform, xen and virtio)
- Docker (podman) (namespaces etc..)
- Selinux and other security bits
- Much more networking options and drivers

@memeka
- Could you please apply the attached patch which fixes the Xu3 analog headphone jack volume? (I can open a PR if you like)

odroid_xu3_headphone_volume_fix.patch.zip

(479 Bytes) Downloaded 204 times

- Did you apply the drm plane fix for kodi? (I believe it was about renaming the cursor plane)

[odroid]

No, I only tried wayland - Weston and Gnome3

I see.
The stock Mate Desktop doesn't start.

[rooted]

I'm running Debian Stretch and the kernel compiled okay, I was getting some odd crashes with gcc when using all 8 cores but after limiting compilation to the big cores it finished.

My ATSC TV tuner and USB 3 hub with integrated gigabit is working, obviously the cloudshell display isn't but I can live with that for now.

I will post up if anything negative happens.

Boot log: https://pastebin.com/i6mj536C

I copied the configuration from 4.14.150 then enabled the new defaults for 5.4.y and all is well.

Configuration: https://pastebin.com/CYC5NTMF

[rooted]

Quick eMMC Test:

Code: Select all

hdparm -tT /dev/mmcblk0

/dev/mmcblk0:
 Timing cached reads:   2240 MB in  2.00 seconds = 1120.97 MB/sec
 Timing buffered disk reads: 488 MB in  3.01 seconds = 162.11 MB/sec

Code: Select all

dd if=/dev/urandom of=test bs=4M count=500
500+0 records in
500+0 records out
2097152000 bytes (2.1 GB, 2.0 GiB) copied, 45.6973 s, 45.9 MB/s

[MastaG]

@rooted, I've gotten similar crashes before on high load (e.g. crashes when compiling).
Be sure to use the odroidxu4_defconfig and then make your changes using menuconfig.
Or just use the config I posted but don't try to use the 4.14 config as a base as that will give problems.
Also make sure to replace the dtb files /media/boot with the ones from the new kernel.

[rooted]

@rooted, I've gotten similar crashes before on high load (e.g. crashes when compiling).
Be sure to use the odroidxu4_defconfig and then make your changes using menuconfig.
Or just use the config I posted but don't try to use the 4.14 config as a base as that will give problems.
Also make sure to replace the dtb files /media/boot with the ones from the new kernel.

Why would it cause problems? I didn't just copy the configuration and compile, I copied it and ran (make oldconfig) which integrates all the new kernel features with the current config and removes spurious options. Of course I copied the new DTB but thanks for the reminder.

It's working fine for my needs which are limited as I don't use HDMI on this particular XU4.

[back2future]

[ btw: on 4.14.x .dtb file for a 5.0.0 kernel, we got here only 1 A15, but all 4 A7 available and with 5.0.0 .dtb all cpus got connected ]

[MastaG]

@rooted because 4.14 and 5.4.0 are a decade away from eachother.
Running oldconfig on 5.4 using the 4.14 config will set some things that can cause problems I guess, also I believe suspend/resume will not work.
Therefore it's best to just configure using the odroidxu4_defconfig which has been fixed by memeka so it works.
Or you can use the one I posted if you want to be more "complete" (e,g. docker, virtualization, selinux, more dvb/usb/network driverfs etc).

Also like @back2future mentions, make sure to use the dtbs from 5.4.0.

[memeka]

@rooted i agree with @MastaG
for example with oldconfig i could not get suspend/resume to work like @MastaG suspected
i had to start with a fresh exynos_defconfig and work my way up
i don't have a good explanation why, maybe because there are new config options that were added and are not selected automatically when doing oldconfig.

[rooted]

If I run into anything I will recompile with a manual reconfiguration to verify its not due to a conflict before reporting a bug.

[hominoid]

The usual methods for controlling the fan do not seem to be working. I get an I/O error when writing /sys/devices/virtual/thermal/thermal_zone0/trip_point_0_temp or any of the other trip points. Does anyone know if the fan can be controlled from userland with this kernel?

[rooted]

The usual methods for controlling the fan do not seem to be working. I get an I/O error when writing /sys/devices/virtual/thermal/thermal_zone0/trip_point_0_temp or any of the other trip points. Does anyone know if the fan can be controlled from userland with this kernel?

You will need to set them to be writeable in the kernel configuration.

Code: Select all

 CONFIG_THERMAL_WRITABLE_TRIPS=y 

[hominoid]

Learned something new, thanks.

[rooted]

Learned something new, thanks.

No problem, it should have been enabled but there are so many kernel options it's easy to miss one.

[tobetter]

I've made Ubuntu images with the kernel from the branch of memeka, these images are boot by the mainline U-boot v2019.10. I've just tested the images if they boots or not, the features you may expect would be missing. Once you have the images on your ODORID-XU4, my build firm will keep building whenever the branch is updated and release the update package.
http://ppa.linuxfactory.or.kr/images/raw/armhf/
username : odroid
password : odroid

The partition layout of the image would be different you know, the first partition where kernel image is stored is not FAT but EXT2 and there is boot.scr instead of boot.ini. Please try if you are interesting.

Nnote: that the images are tested with an uSD card, not eMMC yet.

[escalade]

Has there been a speed regression with the latest rebase? Getting terrible scp speeds (around 5MB/s compared to 18MB/s previously) at the moment.

[odroid]

@memeka,
Can you please make a kernel config or/and dt file for Mate/X11 desktop?
Since I think our next Ubuntu 20.04 LTS still will use the Mate/X11 as a default DE, we need another Kernel settings.

[rooted]

Has there been a speed regression with the latest rebase? Getting terrible scp speeds (around 5MB/s compared to 18MB/s previously) at the moment.

I'm getting 39.5 MB/s over SCP from an XU4 running 5.4.0 to another XU4 running 4.14.150, copying from the 4.14.150 device to the 5.4.0 my speeds are the same 39.5 MB/s so something is wrong with your client.

On my phone over AC wireless I only get 3.5 MB/s or so to the XU4 so it's highly dependent on the device your using.

[rooted]

@memeka,
Can you please make a kernel config or/and dt file for Mate/X11 desktop?
Since I think our next Ubuntu 20.04 LTS still will use the Mate/X11 as a default DE, we need another Kernel settings.

Could you try the config I linked above and see if you have the same problem?

https://pastebin.com/CYC5NTMF

I'm not saying use it as the release configuration, just a test.

[joshua.yang]

I can see the GUI Mate/X11 desktop screen using @rooted's .config file.
It can't build it as it is, but it can after some changes related to firmware things. Changed to the below.

Code: Select all

#
# Generic Driver Options
#
CONFIG_UEVENT_HELPER=y
CONFIG_UEVENT_HELPER_PATH=""
CONFIG_DEVTMPFS=y
CONFIG_DEVTMPFS_MOUNT=y
CONFIG_STANDALONE=y
CONFIG_PREVENT_FIRMWARE_BUILD=y

#
# Firmware loader
#
CONFIG_FW_LOADER=y
CONFIG_EXTRA_FIRMWARE=""
#CONFIG_EXTRA_FIRMWARE="edid/480x320.bin edid/1920x1080_24hz.bin edid/1920x1080_23_976hz.bin edid/848x480.bin edid/1152x864_75hz.bin edid/1280x768.bin edid/1920x800.bin edid/1400x1050.bin edid/1792x1344.bin edid/1024x768.bin edid/1280x800.bin edid/1920x1080.bin edid/720x480.bin edid/800x600.bin edid/800x480.bin edid/480x800.bin edid/1280x1024.bin edid/1280x720.bin edid/1440x900.bin edid/1680x1050.bin edid/1360x768.bin edid/1366x768.bin edid/1600x1200.bin edid/1600x900.bin edid/1920x1200_30hz.bin edid/1920x1200_60hz.bin edid/1920x1080_50hz.bin edid/640x480.bin edid/720x576.bin edid/1024x600.bin edid/720x1280_61.bin edid/480x272.bin edid/1080x1920_55.bin"
#CONFIG_EXTRA_FIRMWARE_DIR="firmware"
# CONFIG_FW_LOADER_USER_HELPER is not set
# CONFIG_FW_LOADER_COMPRESS is not set
# end of Firmware loader

I attached the tiny modified config file and a screenshot. Remove .txt from the attached defconfig file before using that.

Thanks.

[rooted]

Yeah sorry about that @joshua.yang, I forgot I copied the various resolution.bin files from the 4.y kernel for full compatibility.

[odroid]

Glad to see that running GPU driver on Mate/X11 desktop with the mainline Kernel 5.4!

[joshua.yang]

Yeah sorry about that @joshua.yang, I forgot I copied the various resolution.bin files from the 4.y kernel for full compatibility.

It's not a big problem
Thanks for the config file!

[rooted]

You're welcome, I'm glad it works with your change

[escalade]

Has there been a speed regression with the latest rebase? Getting terrible scp speeds (around 5MB/s compared to 18MB/s previously) at the moment.

I'm getting 39.5 MB/s over SCP from an XU4 running 5.4.0 to another XU4 running 4.14.150, copying from the 4.14.150 device to the 5.4.0 my speeds are the same 39.5 MB/s so something is wrong with your client.

On my phone over AC wireless I only get 3.5 MB/s or so to the XU4 so it's highly dependent on the device your using.

Clients are Macbook Pro And NUC8, both Core i5

This is over wireless though, haven't tested cable as my XU4 is inside a bartop. I did a test over localhost and speed was blazing fast so I suppose there's no speed regression. Anybody else seeing poor wireless speeds? I'm using the Odroid wifi module 5A with RTL8812AU driver.

[odroid]

I think we have to merge the vendors' wireless drivers since the mainline kernel wlan drivers usually have some performance/stability/compatibility issues.

We will implement the SPI and PWM drivers first to support the CloudShell2 and OGST LCD display features first.
And later, we will port the wireless drivers.

[memeka]

Sorry I was busy in the weekend with my kid's 1st birthday!
So i missed the conversation.

I will add another patch for mali soon for 5.4.1 rebase, and the changes proposed here.

For mali/x11 there is nothing wrong with the driver, it must be a config issue.
@rooted: what changes did you make?

I am more worried about the SD performance.
With class 10 sd-card i get 2mb/s extremely slow! and with UHS i can get 24mb/s (write performance).

Can someone else post some sd card benchmarks?

[odroid]

Did you run trim command before testing the write performance?
How about read performance?

Anyway, we will try to find an old Class-6 SD card for comparing.

P/S: Happy birthday to your youngest son.

[mad_ady]

Happy birthday, indeed!

[rooted]

Sorry I was busy in the weekend with my kid's 1st birthday!
So i missed the conversation.

I will add another patch for mali soon for 5.4.1 rebase, and the changes proposed here.

For mali/x11 there is nothing wrong with the driver, it must be a config issue.
@rooted: what changes did you make?

I am more worried about the SD performance.
With class 10 sd-card i get 2mb/s extremely slow! and with UHS i can get 24mb/s (write performance).

Can someone else post some sd card benchmarks?

It's the same configuration as 4.y with all the defaults of 5.y, necessary for the newer kernel to run.

Happy Birthday to your child, a very special day and it's great you got to be there.

[hominoid]

Can someone else post some sd card benchmarks?

This is a 32GB SanDisk Ultra

Code: Select all

hominoid@xu4-lab:~$ uname -a
Linux xu4-lab 5.4.0+ #2 SMP Sat Nov 30 23:09:24 EST 2019 armv7l armv7l armv7l GNU/Linux

hominoid@xu4-lab:~$ time sh -c "dd if=/dev/zero of=ddfile bs=8k count=250000 && sync"; rm ddfile
250000+0 records in
250000+0 records out
2048000000 bytes (2.0 GB, 1.9 GiB) copied, 93.7846 s, 21.8 MB/s

real	1m38.407s
user	0m0.437s
sys	0m25.387s

[escalade]

Panfrost is getting closer: https://cgit.freedesktop.org/mesa/mesa/ ... 5618a7de41

I'll have a jab at enabling T628 in Mesa tonight, my previous attempts with panfrost have been unsuccessful as Mesa complains drmbusGetid returns "null". Looks like it may just need to be enabled

[rooted]

Panfrost is getting closer: https://cgit.freedesktop.org/mesa/mesa/ ... 5618a7de41

I'll have a jab at enabling T628 in Mesa tonight, my previous attempts with panfrost have been unsuccessful as Mesa complains drmbusGetid returns "null". Looks like it may just need to be enabled

Keep us posted if you get something going