arter97 kernel for Galaxy Note 5

arter97 kernel for Galaxy S6 / S6 edge / Note 5 / S6 edge Plus​

/* Details */

Upgraded to the latest Linux 3.10
Latest Linaro LSK kernel fully merged
Based on the latest Galaxy S6 / Note 5 kernel sources
Fading LED by AndreiLux and other Note 4 developers
Proper, full support for f2fs & backported from latest stable Linux
CPU undervolt support with Synapse
Display color switcher implemented
- Original code from gokhanmoral's Siyah kernel for i9300
- Toggle by quadruply tapping the multitasking button
- This may not work on edge variants

--- Performance & responsiveness improvements ---
UI lags fixed with binder patch
Full unaligned access enabled
Processor scheduler improvements from AndreiLux
Uses LZ4 with zswap
Dynamic fsync from faux123
Networking speed improvements
Random driver backported from mainline Linux
Westwood as default TCP network congestion control
Built with latest Linaro GCC toolchain with latest GNU linker
Entropy hook on storage removed
Default file-system mount option with noatime
ROW I/O scheduler added and set to default
Scheduler, workqueue, network drivers, file-system drivers, commits properly ported from mainline Linux and CodeAurora(Qualcomm)
Interactive CPU governor replaced with CodeAurora(Qualcomm)'s one
Reduced screen-on delay with boosted CPU cores on resume

--- Memory efficiency & management improvements ---
LowMemoryKiller source-code revamped
Various memory tweaks applied
Memory-management commits properly ported from mainline Linux and CodeAurora(Qualcomm)

--- Battery life improvements ---
Intelligently moves apps to LITTLE cores
Power-efficient workqueues from Linaro enabled
Interactive CPU governor patched for better battery life
NVIDIA's power-efficiency improvement commits applied
Power-managements commits properly ported from mainline Linux and CodeAurora(Qualcomm)

And many others... please read the changelog

/* Switch to f2fs */

As mentioned above, this kernel fully supports f2fs.
However, you are completely free to choose not to use f2fs.
You can selectively convert partitions to f2fs using TWRP provided in the downloads.
You should use the provided TWRP at all times after you switched to f2fs.

/* Disclaimer */

Your warranty is now void.
I am not responsible for bricked devices, dead SD cards,
thermonuclear war, or you getting fired because the alarm app failed. Please
do some research if you have any concerns about features included in this kernel
before flashing it! YOU are choosing to make these modifications, and if
you point the finger at me for messing up your device, I will laugh at you. Hard. A lot.

/* Supported devices */

--- Galaxy S6---
SM-G920F
SM-G920I
SM-G920S
SM-G920K
SM-G920L
SM-G920T
SM-G920W8

--- Galaxy S6 edge---
SM-G925F
SM-G925I
SM-G925S
SM-G925K
SM-G925L
SM-G925T
SM-G925W8

--- Galaxy Note 5 ---
SM-N920C
SM-N920I
SM-N920G
SM-N920S
SM-N920K
SM-N920L
SM-N920T
SM-N920W8

--- Galaxy S6 edge Plus ---
SM-G928F
SM-G928C
SM-G928I
SM-G928G
SM-G928S
SM-G928K
SM-G928L
SM-G928T
SM-G928W8

/* Warning */

Redistribution, modifying files used within this project's file or integrating with other projects are prohibited with no exceptions other than my projects.
Making mirrors, re-uploading to another servers are also prohibited with no exceptions.

/* Downloads and links */

I cannot test this kernel with TWRP.
Switching to the provided PhilZ Touch recovery is highly recommended for all users.

arter97.com

/* Thanks to */

AndreiLux
faux123
Linaro
MinL(xens0117)
hsk from Matcl
CodeAurora

/* Contact */

Email : qkrwngud825@gmail.com
Twitter : @arter97_dev

/* Known issues */

None

/* Installation */

1. Put the downloaded kernel on your internal storage
2. Enter ClockworkMod(or PhilZ) recovery
3. Flash kernel

XDA:DevDB Information
arter97 kernel for Galaxy Note 5, Kernel for the Samsung Galaxy Note 5

Contributors
arter97, AndreiLux
Kernel Special Features:

Version Information
Status: Stable

Created 2015-08-30
Last Updated 2016-06-19

/* Changelog */

22.0 - EOL
Support for future firmwares are not guaranteed!
Fixed support for systemless SuperSU
Latest G920FXXU4DPGW sources merged
Linux 4.8 random drivers merged(12 times faster)
- Please remove Seeder if installed as it's completely unnecessary
TWRP rebuilt with latest f2fs-tools v1.7.0
I/O scheduler optimization for UFS(nr_requests)
Added support for Note 7(Grace UI) ROMs
- Please download the right kernel; it's in another directory
- Incomplete support. Few features might be broken as Samsung have not released the sources.

21.1 - EOL
Support for future firmwares are not guaranteed!
Added support for systemless SuperSU
ADB daemon replaced to stock ones
ADB switched to secure mode
TWRP rebuilt with stock adbd for better compatibility
Galaxy S6 flat - LDU correction disabled

21.0
Memory managements sources related to compression merged with Galaxy S7
Swappiness set to 190 for Galaxy S6
Reverted HMP scheduler commits that causes random kernel panic
Enabled HW accelerated SHA2, GHASH, AES_CCM, AES_NEON_BLK
Reverted cpuidle commits to improve performance
Recovery rebuilt to fix adb sideload, f2fs conversion
Fixed Synapse error for Galaxy Note 5, S6 edge Plus
Support added for Xbox game controller

20.0
Major memory leak fixed
Implemented new feature that moves background applications to LITTLE cores automatically to save power
- Default is set to move unimportant apps to 1 LITTLE core(cpu0)
- You can use Synapse to make it more aggressive - move all background apps(including apps like music player) to LITTLE cores(cpu0~3)
Applied upstream f2fs bug fixes
Revamped CPU governor to fix misc issues due to improper merges
ren_max_freq renamed to scaling_max_freq to fix apps like Kernel Aduitor
DVFS auto resetter implemented to workaround ROM's DVFS race-condition resulting in 1200 MHz clock stuck
Removed dynamic fsync for better reliability and memory efficiency
Re-applied previous Mali GPU commits (by AndreiLux)
- Fixes voltage control
- Adjusts throttling threshold
Merged latest memory management behaviors from Nexus 6P Android N and Galaxy S6 stock ROM
Latest Samsung kernel sources merged
Re-introduced sound-control by AndreiLux as a separate release for Galaxy S6
- You may encounter in-call audio issues with this
- Install the regular version if you don't need this
SysV IPC removed
LOAD_FREQ adjustment hack removed
Kernel tick clock changed from 100 Hz to 300 Hz for better performance
Memory management commits from Nexus 6P Android N merged
Few commits and hacks attempted to reduce resume time
Stand-by ARP and multicast packet filters enabled on Wi-Fi drivers (by AndreiLux)
Built with Linaro GCC 6.1
Binder replaced with CodeAurora sources used with Snapdragon 810
ADB updated
Support added for SM-G920T/W8
Support added for SM-G925T/W8
Support added for SM-N920C/I/G/S/K/L/T/W8
Support added for SM-G928F/C/I/G/S/K/L/T/W8

13.1
Updated to Linux 3.10.101
f2fs updated to Linux 4.7 merge window
Music studderings while screen-off fixed
DriveDroid support added
Synapse fixed
Few memory optimizations
TWRP recovery updated to 3.0.2-0

13.0
Updated to Linux 3.10.100
f2fs updated to Linux 4.6 merge window
Removed improper OOM operations resulting in lags
Merged some optimizations from Liquorix kernel
vnswap updated to Galaxy S7 kernel
Few arm64 optimizations added from upstream Linux kernel
Re-merged HMP scheduler updates from Linaro LSK kernel
Fixed OTG and MTP on TWRP recovery

12.0
Updated to Linux 3.10.98
Support for Android 6.0 added
Support for Android 5.1 dropped
- You will encounter boot issues if you're not properly on Marshmallow firmware
Switched to TWRP recovery
- Make sure to use SuperSU v2.68+ if you want to root your device
Switched to Linaro toolchain 5.3 with GNU linker 2.26
f2fs updated to Linux 4.5 merge window
Adaptive LMK disabled by default
SLUB memory allocator performance improved
Insecure ADB from Android N merged
- Make sure to disable USB debugging if you don't use that feature
zswap, zsmalloc(used with vnswap) merged from latest upstream Linux
Memory managements tweaked
interactive CPU governor tweaked for better responsiveness
Note 5 - used GPU configurations from the S6

11.0
Updated to Linux 3.10.94
Latest Note 5, S6 kernel sources merged
Latest ramdisk, dtb.img merged
Slow Wi-Fi turn-on issue fixed with memory tweaks

X (10.0)
https://en.wikipedia.org/wiki/Roman_numerals
Few possible binder errors fixed
Minor compiler optimizations done to binder
New linker used with compilation (hg-binutils 2.25.51.0.4)

10.0-beta2
Samsung framework's automatic application killer(SPCM) re-enabled
Adaptive LMK re-enabled
scaling_max_freq renamed to ren_max_freq
- Please use Synapse to change CPU max frequency
Core limit on screen-off removed
- Careful observation shows there's not much of a difference
Responsiveness and power consumption tweaks made to hotplugging driver

10.0-beta1
Stand-by ARP and multicast packet filters enabled on Wi-Fi drivers
- This may potentially help battery drain on Wi-Fi connected
CPU max frequency override blocked
Samsung framework's automatic application killer(SPCM) disabled
All A53 cores are now allowed to be turned on during screen-off
Few commits reverted to fix screen-on delay
Adaptive LMK disabled to improve multitasking and potentially increase battery life
Fixed an issue with bottom keys unusable when external output is connected

9.4
Updated to Linux 3.10.93
10.0-alpha discontinued after confirming zram's performance is worse than zswap+vnswap
10.0's display color switcher implemented
- Original code from gokhanmoral's Siyah kernel for i9300
- Toggle by quadruply tapping the multitasking button
- This may not work on edge variants
Added option to allow A57 cores to be turned on during screen-off
(Check the OP's reserved post for details)
Possible race-condition during boot fixed
Memory reclaiming function completely removed to fix performance degradation as time goes on

10.0-alpha3
9.3 changes merged
A new display's color switcher implemented
- Original code from gokhanmoral's Siyah kernel for i9300
- Toggle by quadruply tapping the multitasking button
- This may not work on Note 5 or S6 edge Plus
9.3
Switched back to Samsung's CPU governor to fix performance weirdness
CPU governor patched to be more responsiveness
Boot time decreased

10.0-alpha2
9.2 changes merged
CPU frequency boosts on hotplugging to improve responsiveness
9.2
CPU governor downgraded to LA.BR.1.3.2 to improve performance
Kernel panics related to process reclaim fixed

10.0-alpha1
f2fs updated to Linux 4.4 merge window
Process reclaiming code from LG G4 added to LowMemoryKiller
Adaptive LMK disabled for better multitasking
zram and its dependencies updated to the mainline Linux
Replaced zswap+vnswap with zram
Swap size reduced from 1280MB to 1024MB
zram can be potentially faster than zswap with parallelized compression - max_comp_streams
If testing doesn't show improved memory managements, zram will be removed
This kernel is unstable and regular end-user should avoid it.
Try this only if you're brave and back-up before you proceed.

9.1
Kies, Smart-Switch issue fixed
Added proploader for customizing kernel parameters

9.0
Major performance degradation issue fixed

9.0-beta7
FAST mode on vnswap disabled to fix random reboots(kernel panic)
CPU governor updated to LA.BF64.1.2.2_rb4.6

9.0-beta5
IRQ, CPU hotplug commits reverted to reduce screen-on delay
Hotplugging reconfigured to drastically reduce screen-on delay
Cortex-A53 errata 843419 fix re-enabled to fix invalid instructions
Possible no-deep-sleep issue fixed

9.0-beta4
"CPU cores boosted on screen-on request" reverted to fix screen-on delay
Exynos 7420 IPA and PMU changes applied to CodeAurora's CPU governor

9.0-beta3
Hotplugging properly disabled during screen-on for saving few CPU cycles
Only 2 A53 cores are allowed to be turned on during screen-off
- This significantly helps saving battery life during screen-off
more than just lowering CPU frequency
CPU cores boosted on screen-on request earlier than stock kernel by hundreds of milliseconds
for reduced screen-on delay
Interactive CPU governor replaced with CodeAurora(Qualcomm)'s one
- This version has many advancements compared to stock version
(Benchmark)

8.1
RCU, IRQ, topology merged with CodeAurora(Qualcomm)
ION: system heap allocation time reduced

8.0
Updated to Linux 3.10.92

8.0-alpha2
Hundreds of file-system drivers, memory-management commits ported from CodeAurora(Qualcomm)
This kernel is unstable and regular end-user should avoid it.
Try this only if you're brave and back-up before you proceed.

8.0-alpha1
Scheduler, workqueue, network drivers, power-managements commits backported from mainline Linux and CodeAurora(Qualcomm)
for better performance, responsiveness and power consumption
(Thanks to XileForce)
This kernel is unstable and regular end-user should avoid it.
Try this only if you're brave and back-up before you proceed.

7.1
Recovery device detection fixed for Note 5, S6 edge Plus variants

7.0
Updated to Linux 3.10.91
Graphics drivers updated
PhilZ Touch recovery updated
- Fixes USB storage unusable after first eject
- Fixes kernel, recovery, EFS backups

7.0-beta2
SELinux completely disabled
- This should greatly increase available memory, however might decrease stability as Touchwiz Android 5.1 is full of hardcoded SELinux stuffs
Dual-Sim model support dropped due to development inconsistencies

7.0-beta1
LowMemoryKiller minfree, vmpressure adjusted for better multitasking
Latest Note 5, S6 kernel sources merged
Latest ramdisk, dtb.img merged
Input booster issue fixed for S6, S6 edge variants
CONFIG_EXYNOS5_DYNAMIC_CPU_HOTPLUG_SLEEP_PREPARE set to 1 for better stand-by battery life
Support for G9208, G9200, G920FD, N9208, N9200, N920CD added (untested)
- Please read the 'Supported devices' section
- These variants are untested, please report if it's not working as intended

7.0-alpha2
LowMemoryKiller durp fixed
Kernel panic log shortened to prevent truncation
This kernel is unstable and regular end-user should avoid it.
Try this only if you're brave and back-up before you proceed.

7.0-alpha1
LowMemoryKiller replaced with CodeAurora(Qualcomm)'s version
LowMemoryKiller optimizations(via RBTREE) by Motorola merged
Adaptive LMK enabled
LMK values updated from Nexus 6
f2fs and other drivers updated to use new shrinker API
Automatic process-based memory reclaim method added and enabled
"Drop file-system caches on screen-off to reduce screen-on delay" reverted
This kernel is unstable and regular end-user should avoid it.
Try this only if you're brave and back-up before you proceed.

6.0
f2fs updated to Linux 4.3-rc4
ROW I/O Scheduler by Qualcomm added and set as the default

5.1
Updated to Linux 3.10.90
TCP Congestion Control set to westwood
N920G and G928G support added

5.0
Updated to Linux 3.10.88
Major memory managements changes for better responsiveness
- Reverted vfs_cache_pressure back to 100
- Dirty writeback adjusted(reduced)
- dirty_expire_centisecs and user_reserve_kbytes tweaked
- Drop file-system caches on screen-off to reduce screen-on delay
- S6 & S6 edge - LowMemoryKiller values tweaked
- Note 5 & S6 edge Plus - LowMemoryKiller values reverted back to stock
- Replaced FHA with DHA
- swap-on-disk disabled by default(/system/swapfile)

4.3
Power efficient workqueues enabled by default
Few optimizations to storage performance
Added option to revert back to stock memory managements
- This will give you even better responsiveness at the cost of vastly reduced multitasking performance and increased chance of background services malfunctioning
- Install use_stock_memory_managements.zip from the recovery to enable
- Remove /system/bin/lmktweak to disable
- This will only work on 4.3+ versions
- This is not enabled by default

4.2
Sound-control temporarily removed for fixing in-call volume

4.1
Potential performance degradation issue fixed
VPNCLIENT_SECURITY enabled
(Fixes Wi-Fi calling on T-Mobile)

4.0
G928F,T,W8 support added
Sound-control by AndreiLux added
Synapse loading fixed
Serious malfunction related to LowMemoryKiller issue fixed
Attempt made to fix T-Mobile Wi-Fi calling (Note 5/S6 edge Plus only, untested)

3.6
N920C,I,S,K,L,T,W8 and G928C,I,S,K,L support added

/* To kang or not to kang */

I prefer "open" to all stuffs, explicitly except for my personal kernels.

I've once used to use private Git repository for my kernels and just release the source-codes as a tarball,
which means that other developers who are interested in my kernel's changes would not get the specific changes they want while not violating XDA forum rules or GPL.

That was almost 3 years ago which by then, I was a newbie developer on XDA.
I was afraid other big developers merging all of my changes and I'll get buried down.
However, I've changed my mind and opened up my Git access to everyone after thinking my previous actions did not belongs to XDA's spirit.
That was a big move for myself of re-thinking what should be open.

Later down on the road as I gained more development skills, I've contributed into CyanogenMod and other's custom kernel and ROMs for fixing stuffs, introducing new concepts or improving performance.
During such process, there were quite a lot of occurrences where I was actually the first one to do those things.
For those who're familiar with "temasek ROM", you may know that after I became a contributor, temasek rose even more into the center of other AOSP ROMs' attention, as I was the first to introduce new exciting stuffs such as compiler changes or f2fs. And I quickly gained dozens of followers on my GitHub account.
And eventually, people *kang*ed my changes. Sometimes invalidating my authorship.
It surely is not a right thing to do, but I let it slide as it didn't matter that much. I believed that if my changes are good enough, it'd be better to end up with a larger user base rather than being a **** trying to hunt down every ROM developers who're doing it wrong.


Same story goes to my ROMs.

My real first popular ROM was back-to-n00t, which is a ported ROM from Galaxy S3 LTE to provide stable Touchwiz KitKat experience to the Galaxy S3 3G users.
I opened up everything to GitHub. And I even encouraged that other ROM developers to use my ROM as a base to work on their own ROM as I knew there are a lot of Touchwiz ROM cookers who're way better than me on modifying SystemUI or frameworks(via smali) BUT aren't too good on porting ROMs. I hoped that my ports can be a great base for others to work on their own modifications.
Now those spirits repeated on the Galaxy S6, I opened up everything about back-to-n0t3.


However, I do not do or believe those same things with the kernels.
There are far less kernel developers on XDA compared to ROM developers. ROMs are much more appealing for an "Android newbie" to go tweak. You can modify UI, enable some hidden settings and even port features from another devices.
So when ROMs gets copied off, users can easily distinguish if that was stolen or not.
Let’s say a well-known developer A ported S6 designs and features to the S4. Next day, less-known developer B uploads a ROM on XDA that claims to do the same thing. First thing that comes to the users’ mind would be ‘Is that guy a thief?’.
However, same thing cannot be said on the kernels as they work on a much lower level. You can’t distinguish individual custom kernels just by staring at your phone.
If a ROM developer choose to go use other’s custom kernel and integrate it into their ROM, it’s most likely that users won’t appreciate that custom kernel developer whose work powers that ROM. Let’s face it, not all users care about “Thanks to”, “Credits” sections, and let me tell you, those are the majority. Now some custom kernel developers might be OK with that. However, I’m not. Unlike most developers on XDA, I got fascinated to Android because it runs on Linux kernel. Modifying Linux kernel was the first thing I’ve done on my Android device, not modifying ROM. And I spent years and years on gaining information on Linux/Android kernels and tweaking, coding myself. While I don’t necessarily care users choosing to NOT use my kernel, I cannot stand that my kernel would be installed on one’s device which its owner don’t even know which kernel, which developer who made that kernel is. That is why I’m against with the idea of integrating custom kernels into custom ROMs. If you want your users to have a custom kernel, let them learn themselves what that kernel really is, what does it offers, and who developed it. “For the ease of flashing/installation” is not an excuse for me. Take yourself a few more minutes to download a kernel from the web, and a few more seconds to flash a secondary ZIP file from the recovery. Is that hard? Time-consuming? Absolutely not.

I hope everyone gets the idea why I'm sticking with the idea of "don't integrate my kernel into others".

Cheers.

kisstom said:

Can you please fix this kernel to gear vr. It's extremly laggy

Click to expand...

Click to collapse

Donate me a Gear VR.

I've done some interesting benchmarks targeting semi-realworld situations which are not measurable with traditional benchmarking apps such as Antutu or PCMark, similar to the I/O scheduler benchmarks I've done earlier but with automated tool this time.

I used adb to automate and measure those 2
1. Repeatedly turning on and off the screen 100 times
2. Open up 25 apps sequentially and repeat it 5 times
And just in case, I've uploaded the scripts I've used to GitHub :
https://github.com/arter97/android_autobench

Versions below 3.0 were excluded as those do not support back-to-n0t3, which I'm currently using.
Versions that doesn't have significant changes in performance were excluded.


5.0 had a feature that returns cached memory to available memory to increase memory efficiency thus decreasing screen-on delay. Turns out it was making it worse, by a significant margin.
Future releases had it improved a bit, and that feature was removed all together on 7.0+.


There are only I/O scheduler changes in 5.0 -> 6.0, which is a switch from FIOPS to ROW.
The test itself was done with all apps and background services killed, so judging by the fact that FIOPS has an advantage over ROW on single-threaded read performance, it explains why, this test case in particular, ROW has a worse score.
In real-world, 6.0 is faster than 5.0(as I demonstrated in earlier I/O scheduler benchmark video). And thus, the second test's result can be perceived as my kernel's application launching speed was increased every time as the version goes up.


This test shows that 8.0-alpha2 is quite solid. (And I'm quite happy about it)
And there's actually one more interesting conclusion.

The ~150 commits applied to 8.0-alpha2 were taken from Qualcomm.
And these commits can be easily found in many S808,810 kernels.
I've found most of these commits on the LG G4. That explains why LG G4 surprisingly has very good real-world performance.
On the contrary, OnePlus 2 doesn't have any those commits as it's based on old Qualcomm's kernel.
That explains why OnePlus 2, despite it has S810 over the slower G4's S808, had much slower performance in this PhoneBuff speed-test(which I personally prefer):
https://www.youtube.com/watch?v=JV8LlTxcIvY

This again proves Samsung lacks in software while Qualcomm does cutting-edge changes like (snap-snap-snap-snap)crazy, stabilizing it and opening it for manufacturer to go apply it.
It's a pity that our devices, which CLEARLY has a HUGE advantage over S808/810 hardware, doesn't perform as well as what we expect from such amazing hardware.