Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image)

Luksus · 20 Feb 2024, 10:40

Hi,

I've created a new build for the Oneplus5/T, which is based on the nowadays for android9+ devices recommended build method (Standalone kernel method).
I am running it myself for some months now and also just fixed the last issue, the not working fingerprint, that I think it's time to let it test by you owners for regressions, compared to the current full-system-image version.

Why a new build method at all, since the current build is working fine?

Reduce CI-load
The full-system-image method is very "ressource-hungry". For every device with full-system-image, our jenkins-CI needs to build a (devel) build every day. For the OP5/T it takes about 7 minutes. That doesn't sound much, but the build-server seems to be very powerful, since a system-image build on my local machine takes about half an hour (Ryzen 5700U Processor). Also, these CI-Pipelines need to be maintained by someone.
So, for the standalone-kernel-method, the CI only needs to build one generic_arm64 image (for every Halium version), which will be used by many devices, and the build of that image only takes about 2 minutes.
Better maintainability
The kernel-standalone-method in combination with the generic halium image is more future-proof, regarding maintainability, and also new features or system-upgrades. The general needed updates for newer Halium-versions are centralized and the same for every device.
A device maintainer only needs to take care for the kernel and the relatively small device-config tree with device-specific overlays.

Other improvements of the new build

It already contains the fix for the microphone not working after a call. It survives system-image upgrades.
It contains an initially disabled user-service, that, when enabled, is switching mobile-data to 2G when the screen is off, to save battery.
You can activate it by executing in the terminal on the phone: systemctl --user enable lte-battery-saver.service
It will be started on next boot or you start it manually by executing: systemctl --user start lte-battery-saver.service.
It survives system-image upgrades.
For device-specific needed system-changes, it is using the overlaystore method. Compared to the current method, it is more reliable when doing system-upgrades. The overlays of the current build, may be overwritten by an OTA update, which can lead to device not booting and you would need to use the ubports-installer to reinstall the system-image. With the overlaystore method, such things should not happen.

Flashing and Testing

You can just flash the boot and system image with fastboot, without loosing any data.
At least flashing the system image should be enough, since we are using the same kernel source for both build-methods.

HINT
For me, my local distributions fastboot version does not work for flashing the system image. Therefor I am using mfastboot, a motorola implementation, which works for me.

Get the latest images here:
https://gitlab.com/ubports/porting/community-ports/android9/oneplus-5/oneplus-cheeseburger-dumpling/-/jobs/artifacts/master/download?job=devel-flashable-focal

Extract it, boot the device in bootloader mode, connect it to your pc and flash the ubuntu.img to system partition: fastboot flash system ubuntu.img.
Then reboot.

Snaps

The rootfs/system-image size of the OP5/T is very limited. That's why it is hard to install anything additional on the rootfs, for example the snap-dependencies.
But it is possible to expand it a little bit. For that:

reboot to recovery
connect through adb
execute: e2fsck -f /dev/block/by-name/system

execute: resize2fs /dev/block/by-name/system
Output should be something like:

Resizing the filesystem on /dev/block/sde21 to 786432 (4k) blocks.
The filesystem on /dev/block/sde21 is now 786432 (4k) blocks long.

reboot

Now it should be possible to install snap-dependencies, using the app Snapz0r.

Finally

Please share your experience here.
Especially, I am interested in feedback for the OnePlus5 (without T), since I don't have this device to test it myself.

danfro · 20 Feb 2024, 15:05

@Luksus Just some followup questions before I try this:

Can it be safely reverted in case it doesn't work out? I am a bit cautious since this would be on my daily driver.
Does it matter which channel I am using? Currently focal devel.
Will it affect (for good or for worse) dual sim usage?

But it is good to see things taking steps forward. I am looking forward to try it.

malditobastardo · 20 Feb 2024, 15:43

i am also willing to give it a try on my OP5, btw, does this also include the necessary flags to be able to run virtualization and winebox?

Thank you for your great work!

Luksus · 20 Feb 2024, 16:39

@danfro
yes, you can just install the official build with the installer, or even by OTA update again.
The channel you are currently on, does not matter. My build is using the devel channel.
Regarding dual sim, I don't know.

Luksus · 20 Feb 2024, 16:40

@malditobastardo it is using the same kernel like the official builds, therefor it should be included as far as I know.

Edit:
So it seems, that the kernel config BINFMT_MISC, which is needed for winebox, is not set.
But as I said, you just can update the system-image and leave your boot-image as it is.

Which configs are necessary for visualization?

danfro · 20 Feb 2024, 21:25

@Luksus Thanks for the quick answers. I flashed the ubuntu.img. Log as follows, especially because of the first error message:

fastboot flash system ubuntu.img
Invalid sparse file format at header magic
Sending sparse 'system' 1/6 (494594 KB)            OKAY [ 20.907s]
Writing 'system'                                   OKAY [  0.009s]
Sending sparse 'system' 2/6 (518366 KB)            OKAY [ 24.441s]
Writing 'system'                                   OKAY [  0.015s]
Sending sparse 'system' 3/6 (504130 KB)            OKAY [ 23.555s]
Writing 'system'                                   OKAY [  0.007s]
Sending sparse 'system' 4/6 (524284 KB)            OKAY [ 24.233s]
Writing 'system'                                   OKAY [  0.005s]
Sending sparse 'system' 5/6 (502723 KB)            OKAY [ 23.165s]
Writing 'system'                                   OKAY [  0.005s]
Sending sparse 'system' 6/6 (82784 KB)             OKAY [ 17.075s]
Writing 'system'                                   OKAY [  0.010s]
Finished. Total time: 137.869s

So far it did reboot, load the system and no obvious issues.
Wireguard kernel patch seems to be present too.
Hardware button on the left side works.
Fingerprint works.

Attaching with usb now brings up the devices name as "OnePlus 5", before it was "OnePlus5T" (edit: or "OP5T" not 100% sure). My device is a OP5T.

Some known issues with dual sim are the same:

no mobile data on slot 2
sim card selected for mobile data is not remembered (although that is more a Lomiri issue, not kernel bases I think)

BUT the good news is, one issue seems to be solved: both of my sim cards get detected on every reboot (done about 10 or so). With the other kernel it would take sometimes up to 4 reboots for both cards to be detected. This is a really great improvement!
:crying_face: after applying the battery saver patch and rebooting, the "one sim card is not detected" issue came up again at a 50:50 rate. Can that be related?

Regarding battery saving, I am going to give this a try. Thinking of battery saver app, can this patch be amended so it also turns to 2G when wifi is on (even with the screen on)?

Extending system partition size worked. Installing the snap support and snaps worked too.

And one last question: I should not do any devel/RC/OTA updates right now since that would overwrite this kernel, right?

Btw. on a footnote, your footnote here in the forum does not list the OP5T.

Luksus · 21 Feb 2024, 00:44

Thanks for the detailed feedback.
As I tried to explain, the kernel and kernel features itself should be the same as in the official image... the difference is the system-image and how device-overlays are getting applied.

Because of that, i.e. the usb-config is more generic and does not allow to distinguish between OP5 and OP5T.

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):

after applying the battery saver patch and rebooting, the "one sim card is not detected" issue came up again at a 50:50 rate. Can that be related?

Maybe, since it is modifying sim cards mobile data config.

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):

can this patch be amended so it also turns to 2G when wifi is on (even with the screen on)?

Sure. I have that on my mind too for a while, but did not find the time to look into it.
Maybe there is already an attempt somewhere in this thread:
https://forums.ubports.com/post/51164

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):

And one last question: I should not do any devel/RC/OTA updates right now since that would overwrite this kernel, right?

Yes, kind of.
I am not sure, if it would replace the kernel, but it would at least switch to the official system build again.

danfro · 21 Feb 2024, 05:21

@Luksus said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):

Thanks for the detailed feedback.
As I tried to explain, the kernel and kernel features itself should be the same as in the official image... the difference is the system-image and how device-overlays are getting applied.

Because of that, i.e. the usb-config is more generic and does not allow to distinguish between OP5 and OP5T.

Just my hope that some things get magically solved so I mentioned or tested those things.

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):
after applying the battery saver patch and rebooting, the "one sim card is not detected" issue came up again at a 50:50 rate. Can that be related?
Maybe, since it is modifying sim cards mobile data config.

I turned the script off later to see if the sim card detection then works again. No. But I also have the impression, that it works good after doing some larger (kernel?) updates/OTA or flashing a kernel manually as here. As soon as I change or update something or whatever things go bad again. I guess I could only test this by flashing the whole thing again.

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):
can this patch be amended so it also turns to 2G when wifi is on (even with the screen on)?
Sure. I have that on my mind too for a while, but did not find the time to look into it.
Maybe there is already an attempt somewhere in this thread:
https://forums.ubports.com/post/51164

Quickly zapping over the post I couldn't see anything regarding wifi. I may give it a more detailed read later. But battery safer app has this already implemented, which I was using. I turned that off for testing this kernel.

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):
And one last question: I should not do any devel/RC/OTA updates right now since that would overwrite this kernel, right?
Yes, kind of.
I am not sure, if it would replace the kernel, but it would at least switch to the official system build again.

I will not update then for a while and see how this version works.

malditobastardo · 21 Feb 2024, 07:02

Just some preliminary report:

I got the same error message as @danfro "invalid sparse file format.." but everything was OK and it flashed fine
the keyboard is smaller??
I've had a bug where I could not write anything and I had to reboot
snaps after the space "hack" finally works again
I can confirm the kernel config BINFMT_MISC is not set as you mentioned, but not biggie since winebox never really worked for me on this device.

I will apply now the battery saver ""patch"" and see if I notice some improvements. I will continue testing and report any findings.

Luksus · 21 Feb 2024, 10:37

@malditobastardo Oh yes, I forgot to mention that:
I've did some adjustments (improvements in my opinion) to the keyboard, that it fits better the display height. Mainly for the landscape mode, but then I found, that it could use an adjustment in portrait mode too.

And I made the font smaller in the the "word-ribbon" (the line above the keyboard, with suggestions or the options to copy and paste when in editing mode).

If you want you can switch it forth an back to compare. It would be needed to make system writable and then edit the file /opt/halium-overlay/usr/share/maliit/plugins/lomiri-keyboard/keys/key_constants.js

The original parts are commented: /* overlay-adjusted */
See: maliit key_constants.js overlay

"invalid sparse file format.."

I have that too, but I don't think its relevant, though I just don't know.

I still wonder, if virtualization (pocketVMs) is working for you?
Do you have any kernel-adjustments for making it work? Which images are you using?

malditobastardo · 27 Feb 2024, 07:32

@Luksus Hello everyone,
After testing this on my OP5 for the past 2 days and experiencing issues with the battery life (it only lasts for 2-3 hours with minimum use), I've decided to rollback to the stable kernel by using the updater (updating from the phone using the channels didn't work).
I will continue running more tests to try to see if this is an isolated case or not.

Luksus · 27 Feb 2024, 09:19

@malditobastardo thanks forums testing.

I wonder, if maybe your CPU is not clocking down, or something like that. I will check the device tree of the OP5, if there are differences to the OP5t in this regard.

danfro · 27 Feb 2024, 10:08

@Luksus said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):

Because of that, i.e. the usb-config is more generic and does not allow to distinguish between OP5 and OP5T.

Btw. would a label like "OnePlus 5/5T" be possible to show that it could be either one?

Luksus · 27 Feb 2024, 10:36

@danfro should be possible.

Luksus · 1 Mar 2024, 16:34

@danfro said in Call for testing a the new build-variant "standalone-kernel-method" (vs. full-system-image):

Btw. would a label like "OnePlus 5/5T" be possible to show that it could be either one?

Its updated now. (you would need to reflash system image with the latest pipeline build)

danfro · 1 Mar 2024, 20:53

@Luksus Great, thanks for adding that.

But when I try to flash the new ubuntu.img, I get the error message:

$ fastboot flash system ubuntu.img
Sending 'system' (2867200 KB)                      FAILED (remote: 'Requested download size is more than max allowed
')
fastboot: error: Command failed

Is that because of the resized partition?

Luksus · 3 Jan 2024, 22:14

@danfro I don't know, why that happens. But I remember this from FP3 in the early porting days...
Try fastboot -S 300M flash system ubuntu.img

danfro · 2 Mar 2024, 17:11

@Luksus Thanks, looks like that has worked. The label is now shown as suggested above.

The log does show different size numbers than on my initial install of this kernel as shown above.

$ fastboot -S 300M flash system ubuntu.img
Invalid sparse file format at header magic
Sending sparse 'system' 1/9 (300225 KB)            OKAY [ 12.544s]
Writing 'system'                                   OKAY [  0.005s]
Sending sparse 'system' 2/9 (301241 KB)            OKAY [ 15.055s]
Writing 'system'                                   OKAY [  0.009s]
Sending sparse 'system' 3/9 (298985 KB)            OKAY [ 14.681s]
Writing 'system'                                   OKAY [  0.005s]
Sending sparse 'system' 4/9 (302652 KB)            OKAY [ 14.525s]
Writing 'system'                                   OKAY [  0.009s]
Sending sparse 'system' 5/9 (281309 KB)            OKAY [ 13.530s]
Writing 'system'                                   OKAY [  0.007s]
Sending sparse 'system' 6/9 (307196 KB)            OKAY [ 16.153s]
Writing 'system'                                   OKAY [  0.011s]
Sending sparse 'system' 7/9 (293496 KB)            OKAY [ 14.079s]
Writing 'system'                                   OKAY [  0.009s]
Sending sparse 'system' 8/9 (294882 KB)            OKAY [ 14.181s]
Writing 'system'                                   OKAY [  0.007s]
Sending sparse 'system' 9/9 (246952 KB)            OKAY [ 12.233s]
Writing 'system'                                   OKAY [  0.006s]
Finished. Total time: 135.474s

danfro · 3 May 2024, 19:13

Btw. after flashing the kernel, my led works for notifications and charging. Also my sim cards get detected.

As soon as I make my device r/w or apply any other change or an update that is not a full flash, this functionality breaks. No idea why though...

Edit: Nope, did not work this time. A shame. I wish we could reproduce better why led and sim detection work sometimes to at least get an idea where to start fixing this.

Luksus · 20 Mar 2024, 09:37

I have a fix for the notification LED, you can test it with build here:
https://gitlab.com/ubports/porting/community-ports/android9/oneplus-5/oneplus-cheeseburger-dumpling/-/commit/76220f525846ad2f6fa7c6fb63f861912ad576e5/pipelines

I am not sure yet, if this fix will be the final solution.
Maybe Jami has an idea, how to make it better, based on my description in the merge request.