[Iñigo Martinez Lasala]

Ruto Inor wrote:

Could it be related to the known SoC bug that is solved with intel_idle.max_cstate=1 ? On a Bay Trail mini PC, this is required, otherwise random hangs occur at a low frequency. There are bug reports that Cherry Trail SoCs suffer from the same problem.

Yes, that is another posibility. Also i915 rc6 state (for GPU sleep). This weekend I will test with both changes.
The bug you are talking about appeared with kernels 3.1X series and was fixed in 4.XX… but if we limit Cstate to 1 I’m not sure if the tablet will enter into deep sleep.

[Iñigo Martinez Lasala]

Ruto Inor wrote:

Iñigo Martinez Lasala wrote:

Yes, I also realized about it. On vacations I didn’t have wifi for most of the time and no locks in ten days. After coming back home, crashes came back again. I’ve also played with delayed boot script (executed 30 seconds after boot) but no improvements. However I realized that in most of crashes both Chrome and Dropbox were involved. I’ve removed both and four days without locks. Chrome in particular was crashing due to “no GPU channels available”. Still investigating, anyway…

Thanks for your efforts So two possibilities:

1. crashes are related to the GPU. The some possibilities
   1. Some default settings have been changed regarding the GPU in this ROM in the prob.build
   2. The 4Gb RAM does interfere with the GPU (since the RAM is shared with the GPU) and the GPU needs another settings not to crash.
   3. In either cases, Chrome uses GPU acceleration. Could you try to disable as much acceleration as possible from within Chrome ?
2. networking. That would be bad. Did you you Chrome without accessing the network during your off-line period ?

What are the changes in the prop.build file between the various roms ?

Hi.

Build.prop between versions are almost equal. This is not the problem.

Diffs between both versions:

• At /system/codec, hotplug script. Useless. You can remove it.
• At /system/codec, gov script. Minor changes. Seed parameters are useless but could affect for some hungs. Removed.
• At /system/codec, surfaceflinger script. Useless. You can remove it.

It’s importan to disable forced GPU acceleration for 2D at System/Developer options. It clearly affects stabilty and do not provide any benefit.

Other changes: removed memory=2048M at kernel boot options. This enable more than 2GB RAM memory and for me is a must.

Now I’m testing with other kernel parameters for i915 kernel module (GPU) like i915.semaphores=1 and others. It takes time to test because hangs are not so frequently as in the past.

I think there are at least to type of problems and by removing some parameters and forced GPU acceleration I’ve solved or mitigated one of them because previosly I suffered hangs once or twice a day and now only each five to ten days. During my vacation period I didn’t use Chrome….

But definitively it’s mainly a GPU problem, not Wifi, I think.

 

[Iñigo Martinez Lasala]

Install only V6.  Clean install option.

Previously you should have installed Windows drivers in order to install new ROM. Follow the guide in first post.

[Iñigo Martinez Lasala]

Yes, I also realized about it. On vacations I didn’t have wifi for most of the time and no locks in ten days. After coming back home, crashes came back again.

I’ve also played with delayed boot script (executed 30 seconds after boot) but no improvements. However I realized that in most of crashes both Chrome and Dropbox were involved. I’ve removed both and four days without locks. Chrome in particular was crashing due to “no GPU channels available”.

Still investigating, anyway…

[Iñigo Martinez Lasala]

Probably NOT. Not at least from Teclast.

However a port from Android x86 or Remix is possible. Hope somebody do it. I don’t have time either for trying…

Windows 10 runs fine and now I’m trying to stabilize mirek v6 ROM. That would be enought for me.

[Iñigo Martinez Lasala]

As said, Android is unstable and although is a more convenient OS for a tablet, it has a lot of things and problems with both stock ROM and latest mirek ROMs.

Windows however runs fine, although I don’t feel comfortable with it and it wastes more battery than Android.

On the other side, however, X98 plus runs both Android and Windows without problems and battery last for up to 8 hours but it’s worse for gaming.

[Iñigo Martinez Lasala]

Ralf Brüggemann wrote:

Iñigo Martinez Lasala wrote:

I’ve removed both Surfaceflinger fix and shutting down two cores on low battery and no problems in six days. v6, of course. Both can be removed with a root explorer without reflashing. I will keep testing until next week and if no problem found I will explain how to proceed or release a system.img image for flashing.

would be nice if you share how to configure it, so i also can test if it’s working

Inside /system/codecs there are two scripts. Both are made by Mirek. You have to use a root explorer (for example ES File Explorer in root mode) in order to deal with them.

First one is looking each two seconds for CPU speed. If it’s at 1.5Ghz, then cores 2 and 3 are enabled. If not, cores 2 and 3 are disabled. It’s questionable if letting two cores at 1.5Ghz is more power friendly than four at 700Mhz, for example. I’ve moved that script elsewhere.

Second one, is looking for sufarceflinger service each two seconds. If broken, it restart this service. This was added by mirek prior v3. I’ve also moved elsewhere and since I didn’t noted any unexpected behaviour I think it’s not needed.

I’m also testing by disabling GPU acceleration at Developer options inside settings menu.

[Iñigo Martinez Lasala]

You only have to reset Android from Security tab in Android settings (Reset to factory default). You CANNOT reinstall Android the same way you do with windows although you can install another Android ROM (Mirek, Remix or whatever if available).

[Iñigo Martinez Lasala]

Using mirek ROM you can completely remove Windows using partition tool. Look at mirek thread.

[Iñigo Martinez Lasala]

Me. Use media creator tool from Microsoft in order to create a bootable USB with Windows 10 HOME (not pro).

Then, execute setup.exe from USB plugged with OTG cable and wait. It will take about an hour and a half or more.

[Iñigo Martinez Lasala]

I’ve removed both Surfaceflinger fix and shutting down two cores on low battery and no problems in six days. v6, of course.

Both can be removed with a root explorer without reflashing. I will keep testing until next week and if no problem found I will explain how to proceed or release a system.img image for flashing.

 

 

[Iñigo Martinez Lasala]

Anonymous wrote:

Iñigo Martinez Lasala wrote:

I finally managed to extract both v3 and v6 boot.img partition (kernel plus boot environment). Basically, they are equal. Almost no modifications between both. However, in v6, mirek removed 2048MB limit that Teclast had imposed without explanation. Only that. So, the problem with locks could be imposed by removing this limit in kernel. I’ve played with 3400M but kernel only accepts 2048 limit. Anyway, I’m testing with some different options in order to see if I can avoid that problem. Keep you informed. By the way, it would be great if I could replace old 3.X series kernel from Teclast with a new 4.XX series, but unfortunately my Android knowledge is not so large and eventually I would need plenty free time. <noscript></noscript>So, don’t wait anything here. <noscript></noscript>

Interesting finding indeed! Maybe v6 could be utilized if it was repacked with memory limit? I think there is also an issue with power save. I’ve been suffering from poor WiFi strength pretty much all the time (temporary fix is to switch off and on WiFi – this lasts for a while) and but when I connected PSU to the tablet WiFi remained better. Didn’t have time to check this thoroughly unfortunately. Are there some tools to disable power save functionalities from certain functions? IIRC, Mirek has been focusing on power save quite a bit – maybe this can be a reason for poor WiFi as well.

Kernel from v3 and v6 the same. Only 4G memory limit changes.

Between system v3 and v6 are more differences, specially some build.prop parameters and disabling two cores under low load. But most of the ROM is basically the same.

I’ve been playing with v6 system and v3 kernel without memory limit and problem remains. Now I’m testing with v6 system but removing some tweaks in order to locate problem source. I hope it’s core disabling or surfaceflinger restart scripts (or a combination of both) because if not it will be probably memory limit removal and that’s bad. There is a huge difference in terms of performance.

[Iñigo Martinez Lasala]

You have to disable Windows 10 driver signature check. Boot windows in recovery mode. There are thousands of pages where this is explained with screenshots. Use Google.

Has de deshabilitar el chequeo de drivers firmados de Windows 10. Arrancando en modo recuperación se puede hacer. Hay muchas webs donde lo explican. Usa Google para buscar un tutorial o vídeo, es muy sencillo.

[Iñigo Martinez Lasala]

English, please. Use Google translator if needed. This is an international forum and you will get more chances to be helped if using English.

Por favor, en ingles. Usa Google Translator si es necesario. El foro es internacional así que tendrás muchas más posibilidades de que te ayuden si usas el ingles.

[Iñigo Martinez Lasala]

Hi everybody.

I am testing a fix to avoid screen corruption problems. 48 hours and no corruption (usually once a day or more).

I will continue testing for the next days. If I find no problems until Friday, I will create a kernel package for upgrade.

[Author]

Posts