[KERNEL] [Y300/G510] Stock Huawei

[Guest]

Due of untraceable performance regression on R/W and in 3D tasks I'll downgrade the kernel to 3.4.67. That's the upgrade level Google's offical kernel got: https://android.googlesource.com/kernel/common/

[Guest xSeed]

G510-0200 with synopsis kernel 17/07/2014 on AOSP MOD B200

 

 

u0_a70@android:/ $ su
root@android:/ # dmesg | grep LCD_DEBUG
<3>[1, swapper/0] [    0.764811] [LCD_DEBUG] enter mipi_cmd_hx8369a_fwvga_init
<3>[372, SurfaceFlinger] [    7.247014] [LCD_DEBUG] leave mipi_hx8369a_fwvga_lcd_on
<3>[29, kworker/u:2] [   95.877014] [LCD_DEBUG] leave mipi_hx8369a_fwvga_lcd_off
<3>[29, kworker/u:2] [  126.936978] [LCD_DEBUG] leave mipi_hx8369a_fwvga_lcd_on
root@android:/ #

[Guest jor1196]

moddingg33k can create a cm10.1 kernel from source? 

No, not repack

[Guest]

Re-packs are not different from my "original kernel". For all ROMs the kernel core (zImage) is always the same. The only difference between those Android versions is the INIT process during boot, which is getting controlled by scripts being contained on the RAMdisk of the kernel image.

[Guest]

G510-0200 with synopsis kernel 17/07/2014 on AOSP MOD B200

 

Thx. The used driver for different G510 models might be the same. Further I assume the same counts for the Y300. Looks like I can rid of another ~100 KB by disabling the other 8 unused drivers.

[Guest porozex]

Badly written APP :P The Author of it should think about re-writing the ZRAM initialization.

developer said this: amount of zram is calculated assuming that every core has a corresponding zram device. If you have 2 cores and 1 zram device you will have half of zram value

[Guest]

Well that's exactly what I've been trying to say all of the time

 

His english is better than my's. He can explain things simpler

 

He shouldn't assume one device per core but rather check the amount of available devices. It's just a small change and more logically. Or did you ever go to a shop and take something without paying, because you "assumed it was was for free" for example?

Edited July 20, 2014 by Guest

[Guest fonz93]

I decided to use SWAP on ext4 partition in SD card for a while, just for curiosity.

 

What are especifically the advantages and disadvantages of SWAP (From your experience)? (should i try overcommit_memory to "0" too?)

Edited July 20, 2014 by fonz93

[Guest]

I don't use physical SWAP to avoid decreasing the longevity of my flash chips (wear levelling).

In case of using SWAP there are 2 parameters which you could consider:

• overcommit_memory (can be set to 0)
• min_free_order_shift (can be reduced to 2 or even 1)

Edited July 20, 2014 by Guest

[Guest fonz93]

 

I don't use physical SWAP to avoid decreasing the longevity of my flash chips (wear levelling).

 

In case of using SWAP there are 2 parameters which you could consider:

• overcommit_memory (can be set to 0)
• nr_page_shift (can be reduced to 2 or even 1)

 

 

Ok, thank you

 

I will not use SWAP on internal SD but on External SD, so, in case i'll broke it i'll lose only about 10-15 Euros xD

 

Where can i find the tuneable nr_page_shift file?

 

 

Edit: I haven't a configured virtual machine so i can't do it my self, in next update, if is not a problem for you, can you compile a kernel version with 7 MB extra on the GPU, iwant to test the opposite that Ceastel did on his kernel (i need to test it for streaming in High Quality in Twitch app), this is how Ceastel did:

 

https://github.com/Ceastel/CeXstel-kernel-msm8x25/blob/fce45d1f2c0922b9944d1947e15d2b1ff3077784/arch/arm/mach-msm/hardware_self_adapt.c

 

if you search for: CONFIG_CEASTEL_RAM_STOCK

 

You'll find many configurations for screen resolution types, i want to test the QHD configuration: mdp_pmem_size = 0x2300000; //35M

 

Is it possible? Thank you

 

 

Edit 2: under my post you'll find a better explanation by H3ROS :D

Edited July 20, 2014 by fonz93

[Guest H3ROS]

Yeah, this was my suggestion to fonz. With the Ceastel kernel he's added different values so that Android gets more free RAM, but that takes away memory from the GPU. The result is that people have more free memory for their apps, but it breaks video playback. Like the 407MB version of the CeXstel kernel can't play Twitch video steams at high quality or medium quality and the 400MB version can't do high quality. The stock version (393MB) can do all quality settings, although high quality can sometime freeze up. If the GPU was given anything between 1MB to 7MB more then high quality playback should work much better.

 

My suggestion to fonz was to change the value to the QHD value (35MB) and then remove 1MB until high quality video playback suffers. Doing this might also allow YouTube to display higher quality videos.

 

0x2300000 = 35MB

0x2200000 = 34MB

0x2100000 = 33MB

0x2000000 = 32MB

0x1F00000 = 31MB

0x1E00000 = 30MB

0x1D00000 = 29MB

0x1C00000 = 28MB

[Guest]

I guess I need to fix this pmem related issue first:

 

pmem: pmem_setup: unable to register pmem driver(pmem) - zero size passed in!
pmem: pmem_setup: unable to register pmem driver(pmem_adsp) - zero size passed in!
pmem: pmem_setup: unable to register pmem driver(pmem_audio) - zero size passed in!

 

My idea was to get fully rid of the pmem kernel module and to convert existing users to ION (if they weren't already converted anyway).

[Guest fonz93]

Maybe needs changes in the ROM side? or not?

[Guest]

No, that's more likely a kernel source fault (present on all stock kernels by default).

 

Since we already got a very low amount of RAM I don't like the idea of having 2 memory allocators fighting for resources at the same time anyway.

[Guest]

I got the feeling that we should stick to kernel version 3.4.0 since it performs best so far. Any oppinions?

[Guest]

@fonz: the option is called: min_free_order_shift

[Guest jor1196]

the repack always bring some bug 

cm10.1 had the bug in the search for devices with Bluetooth

[Guest fonz93]

I think the overcommit_memory to 0 is made to work well with swap or zram only, now that i have swap is working perfectly but when i tried it 1 week ago without swap i had many problems of non opening apps and others random.

 

 

Edit: I did some antutu benchmarks, i got between 7700 and 8000, i have SWAP 128 MB on Ext4 partition in micro SD card, overcommit_memory=0 and min_free_order_shift=2, my ROM is AOSP MOD by H3ROS and synopsis kernel 17-07.

 

 

Here an archive with all my scores in .txt, i did all tests this morning: http://www.mediafire.com/download/ctu5iuoc099db22/Scores.zip

 

 

This is the maximum i reached: 

 

 

Edited July 21, 2014 by fonz93

[Guest SammichPG]

I'd really like a version with more free ram for general apps at the expense of vram if the performance on the gui and anything non gaming related would remain the same.

 

I don't watch any video or play any game on this phone and having free ram without any disadvantage for my use case would be really good.

[Guest porozex]

I think the overcommit_memory to 0 is made to work well with swap or zram only, now that i have swap is working perfectly but when i tried it 1 week ago without swap i had many problems of non opening apps and others random.

 

 

Edit: I did some antutu benchmarks, i got between 7700 and 8000, i have SWAP 128 MB on Ext4 partition in micro SD card, overcommit_memory=0 and min_free_order_shift=2, my ROM is AOSP MOD by H3ROS and synopsis kernel 17-07.

 

 

Here an archive with all my scores in .txt, i did all tests this morning: http://www.mediafire.com/download/ctu5iuoc099db22/Scores.zip

 

 

This is the maximum i reached: 

 

....

it can be higher i reach around 1500 in 3d test sometimes

[Guest SH3H1]

I got the feeling that we should stick to kernel version 3.4.0 since it performs best so far. Any oppinions?

Sure ! That might be the kernel which can squeez the Y300 more :D

I think sticking with it will make things simpler for ya....

So..,Go if you think it's better :)

[Guest ZolaIII]

Fonz put back min_free_order_shift=4 decreasing it just makes bigger pressure on cache reclaiming totally unnecessary.

My Antutu 3D tests actually score & over 1700...

You are actually getting a little more ram with every unneeded driver disabled & cleaned. Overcoming memory will just make like you have more ram bat you can use it only with enabled swap. & it's more about virtual memory (vmalloc) then real one!

[Guest]

Fonz put back min_free_order_shift=4 decreasing it just makes bigger pressure on cache reclaiming totally unnecessary

 

Maybe a good time to point out that this tuneable only exists an Android kernels and never got integrated into the general Linux kernel. Linux's default is always "1". It got introduced on Android, because usually you don't have SWAP/ZRAM on it.

 

See here: https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/kernel/sysctl.c?h=msm-3.4&id=83de731ffcc6777a33e8a6132c7da8d91faac9ca

[Guest fonz93]

Maybe a good time to point out that this tuneable only exists an Android kernels and never got integrated into the general Linux kernel. Linux's default is always "1". It got introduced on Android, because usually you don't have SWAP/ZRAM on it.

 

See here: https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/kernel/sysctl.c?h=msm-3.4&id=83de731ffcc6777a33e8a6132c7da8d91faac9ca

 

So it's better to decrease it or use the default value?

[Guest]

I'd really like a version with more free ram for general apps at the expense of vram if the performance on the gui and anything non gaming related would remain the same.

 

Did you already tune the values for the OutOfMemoryKiller?

 

Here's some detailed explanation for this topic: http://www.androidcentral.com/fine-tuning-minfree-settings-improving-androids-multi-tasking (just like: http://ajqi.com/android-memory-management-and-its-effects-on-multi-tasking/)