Antutu 3D benchmark & other Unity-based titles cause DeadSystemException & soft-reboot

[Robert Coldbird]

I've been trying to figure out this bug for a while now.
In the 7.1.2-rc2 release iso this issue doesn't exist... but every source-built iso suffers from this issue.

This isn't just limited to android-x86 either, forks like BlissRom-x86 suffer from this as well.
Furthermore, the lack of a 7.1.2-rc2 release tag in the source make it very hard to pinpoint when exactly this issue first came to be.

Can anyone else verify this and maybe share some additional information if they have any?

[fguy]

This error occurs if "ro.hardware.sensor=kbd", if you assign "hdaps" or "iio", then these applications run without problems.

[Robert Coldbird]

Wouldn't that cause issues further down the road, given the fact that my device has no accelerometer?

[fguy]

Wouldn't that cause issues further down the road, given the fact that my device has no accelerometer?

This problem is caused by incorrect processing of the option (property) "config.override_forced_orient" in android 7.1, and in 8.1 there is no such problem.

Disabling (or enabling) this option will result in applications that only work in portrait mode will unfold the picture on the display. 

[Robert Coldbird]

Does anyone know when this issue was first introduced? Or rather, by what commit?
I know that the 7.1.2-rc2 release didn't suffer from this issue.

[Robert Coldbird]

Also, toggling that option changes nothing for me (sadly).

Am Donnerstag, 25. Januar 2018 18:28:40 UTC+1 schrieb fguy:

[fguy]

Also, toggling that option changes nothing for me (sadly).

https://groups.google.com/d/msg/android-x86/efgnq9xA5aQ/9qaU4F7PAwAJ 

[Robert Coldbird]

I've tried all of the points mentioned by you.
Reverting the framework patch, messing with the configuration flags, etc.

However, Antutu and every Unity-based title still fail to launch, and cause the DeadSystemException + followed soft-reboot.

Running out of ideas here... are you sure this is caused by what you say?

[Chih-Wei Huang]

2018-01-26 6:57 GMT+08:00 Robert Coldbird <gentlema...@gmail.com>:
> I've tried all of the points mentioned by you.
> Reverting the framework patch, messing with the configuration flags, etc.
>
> However, Antutu and every Unity-based title still fail to launch, and cause
> the DeadSystemException + followed soft-reboot.
>
> Running out of ideas here... are you sure this is caused by what you say?

Hmm..? I have no problem to run
Antutu 3D benchmark on Surface 3
with the latest image.

What's your GPU?

--
Chih-Wei
Android-x86 project
http://www.android-x86.org

[fguy]

Hmm..? I have no problem to run
Antutu 3D benchmark on Surface 3
with the latest image.

What's your GPU?

  

This problem is not related to the GPU. I observe it both in vmware and in intel 6Y30. It occurs when the kernel does not detect the sensors hub and "init.sh" sets the parameter "ro.hardware.sensor = kbd". 

[Robert Coldbird]

The latest "build" posted on the website isn't affected, that's why you don't see any issues.
The published 7.1.2-rc2 iso still works fine... it is builds after that (fresh ones built from source) that suffer from this issue, which means it must be something that got introduced after the 7.1.2-rc2 iso on the website was built.

[Chih-Wei Huang]

Why did you think it's related to sensors?
The forced_orient function has already in 7.1-rc2
which doesn't suffer from the problem as Robert said.

[Chih-Wei Huang]

2018-01-26 17:57 GMT+08:00 Robert Coldbird <gentlema...@gmail.com>:
> The latest "build" posted on the website isn't affected, that's why you
> don't see any issues.
> The published 7.1.2-rc2 iso still works fine... it is builds after that
> (fresh ones built from source) that suffer from this issue, which means it
> must be something that got introduced after the 7.1.2-rc2 iso on the website
> was built.

I said I tested "latest image" which is built
from the latest code, of course.
It's what I always test.

[Chih-Wei Huang]

2018-01-26 18:28 GMT+08:00 Chih-Wei Huang <cwh...@android-x86.org>:
> 2018-01-26 17:49 GMT+08:00 fguy <fguy...@gmail.com>:
>>> Hmm..? I have no problem to run
>>> Antutu 3D benchmark on Surface 3
>>> with the latest image.
>>> What's your GPU?
>>
>> This problem is not related to the GPU. I observe it both in vmware and in
>> intel 6Y30. It occurs when the kernel does not detect the sensors hub and
>> "init.sh" sets the parameter "ro.hardware.sensor = kbd".
>
> Why did you think it's related to sensors?
> The forced_orient function has already in 7.1-rc2
> which doesn't suffer from the problem as Robert said.

I may reply too quickly. Sorry!

After testing on a device with kbdsensor,
I do see system_server crashing in
libsensorservice.so.

But the root cause is not due to the forced_orient function
but in the recently updated libsensors (beyond 7.1-rc2).

The reason I didn't see it in Surface 3 is because
it uses another sensors HAL (iio).

I'll work out a fix.
Thank you for the report!

[fguy]

I may reply too quickly. Sorry!

After testing on a device with kbdsensor,
I do see system_server crashing in
libsensorservice.so.

But the root cause is not due to the forced_orient function
but in the recently updated libsensors (beyond 7.1-rc2).

The reason I didn't see it in Surface 3 is because
it uses another sensors HAL (iio).

I'll work out a fix.
Thank you for the report!

I'm glad that you could repeat the mistake and hope for its correction.

There is one more issue in the same topic. 

I patched the core and my tablet saw the rotation sensor "iio" (CONFIG_INTEL_ISH_HID=y). 

Unfortunately, this led to the fact that the portrait applications began to unfold. 

Is it possible to save the launch mode of portrait programs in landscape mode if there are sensors? 

I would like that it could be included in the Settings without rebooting the device.

[Robert Coldbird]

If its any help, the oreo-x86 branch is unaffected.

[Robert Coldbird]

I'm not entirely sure it was a change in libsensors that caused this issue...
This issue existed before the r36 merge.

I know it did because I first discovered this issue way before you merged r36 in.

Am Freitag, 26. Januar 2018 11:44:46 UTC+1 schrieb Chih-Wei Huang:

[fguy]

Does anyone know when this issue was first introduced? Or rather, by what commit?
I know that the 7.1.2-rc2 release didn't suffer from this issue.

Now for normal operation of android 7.1.2 on devices without detectable motion sensors, you must cancel the patch https://osdn.net/projects/android-x86/scm/git/hardware-libsensors/commits/1f9a8a2e326c4e4c89ac351084bbc123b19715c0

For already assembled android x86 7.1.2 you need to fix the file "systems/etc/init.sh" in the function init_hal_sensors change:

*)

has_sensors=false

hal_sensors=hdaps

;;

esac

[fguy]

Another variant of the patch for ready assemblies is to use files system\lib\hw\sensors.kbd.so and system\lib64\hw\sensors.kbd.so from 7.1-rc2

[Chih-Wei Huang]

2018-01-27 1:39 GMT+08:00 Robert Coldbird <gentlema...@gmail.com>:
> I'm not entirely sure it was a change in libsensors that caused this
> issue...
> This issue existed before the r36 merge.
> I know it did because I first discovered this issue way before you merged
> r36 in.

Nothing to do with r36.

You can sync the latest hardware/libsensors
which has fixed the crashing.