[X]Assistant

[Alesaiko]

Google Nexus 5 - Kernels and their discussion
LG Nexus 5 D820 / 821 (Hammerhead)
Description | Discussion пїЅ | Purchase пїЅ | Accessories пїЅ | Marriage пїЅ | Nexus 5 Owners Club пїЅ | power usage пїЅ | Benchmark - results пїЅ | Official firmware пїЅ | Unofficial firmware пїЅ | LineageOS пїЅ | Ubuntu пїЅ | MIUI пїЅ | Kernels and their discussion | Modification of software and decorations пїЅ

Important information and instructions

Before any actions with the device, do not forget to backup the system and EFS!

Fiks Wi-Fi(New firmware module wireless networks)
Qualcomm Experimental Proprietary Libraries from Qualcomm

Kernel installation

Before flashing, be sure to verify the type of kernel being flashed!
CAF kernels are designed for CAF firmware. AOSP kernels - for AOSP firmware, respectively.

Preparing to work with Fastboot

Literal way to install the kernel using TWRP:

• Reboot device to recovery (TWRP)
  
• Click "Install"
  
• In the opened explorer find and click on the zip-archive with the kernel. If the installed kernel is represented as a boot.img image, switch the explorer to img mode beforehand.
  
• Install the kernel by pulling the install slider
  
• Reboot to the system

Literal way to install the kernel using Fastboot:

• Prepare the folder with the necessary tools (ADB / Fastboot)
  
• Place the kernel image in .img format in this folder
  
• Reboot device in APPSBL (Fastboot) - Volume down + Power
  
• Make sure the APPSBL (external bootloader) is unlocked.
  
• Sew kernel using terminal command
  fastboot flash boot<core_name>.img
  
  
• Reboot to the system

Do not forget! The core should be stitchedafterinstall the firmware! The bottom line is that each firmware, not counting some incremental OTA updates, comes with its own sectionboot. As the name implies, boot is the boot partition. Roughly speaking, boot is the core, but at the same time not quite. On our device, this block consists of two separate parts: a compressed image of the kernel merged with the configuration tree (DT) - zImage-dtb; and root boot sector - initramfs. Many modern kernels use a special installation scheme: their scripts first disassemble the boot that is already present in the system, separate the initramfs from it and automatically compile the new boot with the replaced zImage-dtb, then the resulting block is flashed. This installation method ensures compatibility with all related firmware of the same Android version, be it LineageOS, stock firmware, or even MIUI.

If, after installing the kernel, the system refuses to start, it means that your kernel is incompatible with the installed firmware, or it contains a critical error. If in the first case it still makes sense to try again, but with a different configuration, then installing a broken kernel is a priori incorrect. Better turn your choice to the other side.

If you have any problems, you can easily return to the stock kernel. To do this, simply flash the boot.img image that comes with your firmware. However, the best way to avoid any problems associated with the tails from the previous kernel is to install the firmware clean, so that the system and boot sections return to their original state.

Kernel Testing and Reviews

Comparative Testing Form

• Testing cores fromimageMaster
  
• The behavior of the cores in the reader Moon Reader fromimageMaster
  
• Franco Kernel: Default Settings vs Battery Saving fromRauchy98
  
• Brain Explosion Review byidmx
  
• Feedback on kernels fromVampir2012
  
• Feedback on kernels fromTysovwikk
  
• Test cores fromVitromanof
  
• Big bum review fromromixer
  
• Comparing kernels fromMarkmaster
  
• Feedback on kernels fromVitromanof
  
• Franco core speed testing fromVitromanof

Kernel Tuning Programs

• Kernel adiutor- an advanced and attractive tweaker with many possibilities.
  
• ElementalX Kernel App- tweaker fromflar2, the author of the kernel ElementalX.
  
• franco.Kernel Updater- tweaker fromfranciscofranco, author kernels Franco Kernel.
  
• Fauxclock- tweaker fromFaux123, the author of the core Faux Kernel and Hybrid.
  
• HellsCore Kernel Manager- tweaker fromHellsgoodHellsCore author.
  
• Synapse- a powerful customizable tweaker for cores with its support.

Kernel settings from our users

• Kernel settings fromaroum
  
• Kernel settings fromTysovwikkandsecond option
  
• Kernel settings fromilmirkin
  
• Kernel settings fromimageMaster
  
• Kernel settings fromindigo88andsecond option
  
• Kernel settings fromUltimateDroid
  
• Kernel settings fromVitromanof
  | Option from March 6, 2016
  | Option from May 23, 2016
  | Option from October 29, 2016
  | Option from February 24, 2017
  | Option from April 5, 2017

Modified Kernels

AOSP kernels for AOSP firmware, CAF kernels for CAF firmware. Be careful before installing the modified kernel!

Actual kernels

AOSP kernels

Breakfast

BREAKFAST

Android 6.0.y - 8.1.0

Speklist

Minimalistic core for AOSP firmware, the concept of which is a minimum of changes relative to the stock core:


• Clean baseandroid-msm-hammerhead-3.4-marshmallow-mr3 (hammerhead)orandroid-msm-flo-3.4-marshmallow-mr2 (flo), official tools and build configuration. The kernel is assembled as the manufacturer intended.
  
  
• The minimum level of intervention in the system. This archive only modifies the block.boot (/dev/block/platform/msm_sdcc.1/by-name/boot)replacingzImage-dtb (compressed kernel image combined with configuration tree)and device-dependent scripts ininitramfson your own. It is very simple to return the original state: it is enough to flash the original boot.img image that comes with the firmware.
  
  
• The complete absence of thousands of useless changes "as long as it was" and heaps of "security patches" for drivers that we do not use. Less than a hundred commits have been made to Breakfast, and most of them are preparation for the compilation, full support for Android N and their own buns. Before including a commit into the kernel, the modified code is carefully analyzed and, if possible, improved to eliminate potential troubles during execution.
  
  
• I completely rewrote the installation script in the shell command shell listing. Due to this, it became possible to more accurately install the kernel from the point of view of theoretical errors in the process of it, and also there appeared complete universality between all supported devices. A sort of personal AnyKernel. In addition, I removed the test for the existence of the power.msm8974.so library (power.msm8960.so in the case of Flo) in the system, since in the presence of a device-dependent analog (power.hammerhead [flo] .so, respectively), this library is not a priori is used. Therefore, if there is a check for a device-dependent PowerHAL, you can ignore the lack of a platform-dependent option. In practice, this will help with installing the kernel on many third-party firmware, the authors of which carefully remove it.
  
  
• The kernel is configured more aggressively in comparison with the drain. The CPU frequency changing policy now raises the frequencies noticeably more readily, while the I / O scheduler is now deadline (at hammerhead it is also used to process requests for data I / O in the mmcblk0rpmb modem unit). I turned my choice in his direction, refusing from CFQ and NO-OP, since the deadline allows you to quickly and simply prioritize read requests relative to those for writing, which is quite relevant in the mobile segment, where random data reading prevails over other I / O operations . Like the NO-OP, the deadline is a FIFO (First-In-First-Out) scheduler, and it also supports front-end occurrences (situations in which an I / O request ignores a queue and flies right in front of it), so that In terms of delays, he will not yield to anyone.
  
  
• I have significantly rewritten the standard CPU frequency changing policies (interactive is not included in this list), making them easier, smarter and easier to read. With the ondemand and conservative, they now more honestly calculate the current workload of the logical processor, and userspace no longer uses its own resources to modify global ones. However, I think technical details are not interesting to anyone (and if interesting, welcome tomy profile on github). I added several new (and not so) features:
  - Conservative henceforth again uses separate stepping values ​​of the maximum frequency to change the target. freq_up_step is used as a percentage of the current (not possible) maximum frequency by which this target frequency will move up. freq_down_step is an analogue for moving down (reducing the frequency). However, this is already familiar to you from my old cores and not only.
  - The sampling_down_factor in conservative now does what it needs for design. Oddly enough, this variable has been broken for four years. Now, as soon as the load on the logical processor grows to the up_threshold_burst threshold value (by the way, this is also a new variable. From this description of sampling_down_factor, you should understand what it does and why), conservative will use sampling_down_factor as a sampling_rate factor,if the previous frequency was lower than the maximum, and will raise the current frequency to the maximum. Obviously, this is done in order to reduce delays during an abrupt jump in load. Multiplied by sampling_down_factor sampling_rate, in turn, significantly reduces the frequency of new snapshots, allowing the processor to operate at an increased frequency of at least sampling_rate * sampling_down_factor milliseconds, which also helps the system to more quickly handle the source of this strong load spike.
  - With ondemand, things are more interesting: the new variable input_boost_freq indicates the frequency to which the current frequency of the logical processor rises during an input event. There should be no problem understanding this variable. Much more significant is the load_scaling variable. Reducing this variable to a non-zero value involves a special formula for calculating the target frequency:
  
  freq_next = min_f + max_load * (max_f - min_f) / 100
  Where:
  - freq_next - target frequency of the logical processor
  - min_f - the minimum possible (non-current) frequency (policy->cpuinfo.min_freq)
  - max_f - the maximum possible (non-current) frequency (policy->cpuinfo.max_freq)
  - max_load is the maximum load among all processors in the cpufreq policy (in our case it is the load of a single logical processor — the core)
  
  At the same time, we should not forget that the usual increase in frequency for ondemand to the maximum while fixing the load above the value of the up_threshold has not gone anywhere. In the future, this formula allows you to use all the available frequencies, effectively neutralizing the predominance of any frequency in the absence of accompanying factors (in our case, PowerHAL and other mechanisms briefly increase the frequencies of all CPUs to certain values ​​in order to improve the response, so that the prevalence of these frequencies will not disappear ), which makes the "swimming" frequency itself much more accurate and flexible. By the way, exactly the same formula is used in Mainline (the latest Linux kernel 4.15+).
  
  
• The kernel contains the infrastructure necessary to correctly calculate the influence of specific tasks on the logical processor, relying on time windows, which ensured the uninterrupted operation of the cpu-boost module, which is used both to speed up the logical processors during input events and during the migration of tasks from one logical processor to another. At the same time, cpu-boost uses data from the above-mentioned infrastructure to preserve the efficiency of CPU acceleration processes during migration. However, cpu-boost is not used by default, since the firmware has a so-called PowerHAL, which performs almost the same tasks from user space. The presence of two simultaneously working solutions that do the same thing, for obvious reasons, leads to a conflict between them. Because of this, I leave cpu-boost inactive on all firmware, except Oreo (which support is not, ha ha), because everything with the interface’s smoothness and responsiveness is very bad (at least on unofficial AICP 13.1). However, you can use cpu-boost on any firmware. To do this, first of all, it is necessary to disable the synchronization of frequencies of logical processors during the migration of tasks in the cpufreq ondemand policy, if it is used, of course. The variable sync_on_migration is responsible for it. You only need to reset this variable if you need to use task migration messages directly from the nuclear scheduler center (only cpu-boost can work with them). The final chain will look something like this:
  
  
  1. First you need to disable mpdecision using the terminal command:
     
     stop mpdecision
     
     Thus, the influence of PowerHAL on the system will be neutralized without the need to interact with the power. *. So libraries themselves. Moreover, using mpdecision with cpu-boost is initially incorrect, as it expects prompts from PowerHAL, which in any case will have to be eliminated.
     
     
  2. Now it is worthwhile to use direct messages about the migration of tasks between logical processors from kernel / sched / core, having previously turned off synchronization of frequencies in ondemand using terminal commands:
     
     echo 0>/ sys / devices / system / cpu / cpufreq / ondemand / sync_on_migration
     echo 1>/dev/cpuctl/cpu.notify_on_migrate
     # Does not exist in Oreo
     echo 1>/dev/cpuctl/bg_non_interactive/cpu.notify_on_migrate
     
     
  3. Finally, it is already possible to activate cpu-boost with the help of a tweaker or manually (but more conveniently, nevertheless, with the help of a tweaker).
     
     If suddenly the desire to use cpu-boost disappears, you need to perform the above steps in reverse order, or simply reboot the device.
  
  
• Clean and minimally crutch implementation of features familiar to you from other custom kernels.
  
  
• The ability to change the values ​​of amplification of sound lines, based on the restriction of the right to write to the target registers from the user space through the IOCTL interface. The fact is that the Android operating system uses the IOCTL interface to change the gain values ​​in accordance with the configuration specified in the configuration file/system/etc/mixer_paths.xml, after each reinitialization of the audio unit. For the user, this means that any changes in the gain values ​​from somewhere outside of AudioHAL will be safely overwritten by the system as soon as the sound device, whose gain has been changed, becomes active again. Many people know this in the form of resetting the module settings.Soundcontrolfrom uvFaux123and that of uv.franciscofranco. I use the easiest and “soft” for the system method of preventing rewriting: the registers whose gain values ​​will be subject to change through the sysfs interface are no longer available for recording via the IOCTL interface. For backward compatibility withSoundcontrolmy module initializes its sysfs outputs to/ sys / kernel / sound_control_3. However, there is a downside:The gain values ​​on the RO line from mixer_paths will no longer have any effect. In simple words, the so-called. The “sound modes” that amplify the speaker / microphone / headset will stop working. I do not think that this will be a problem for anyone, since the nuclear option offers the same opportunity, only working in real time.
  
  
• The minimum set of changes to generate the interface used to control the color scheme of the panel. Unlike other kernels, my options are noticeably lighter and are written in better quality language. So, my optionKCALlighter than the original from SW.Savocaby 61% andinterface to the registers payloadIn the MDSS controller, the lighter is the original from the SW.Myfluxiby 93%. Both interfaces are written using the best syntax present in our old kernel.
  
  
• Since, in fact, the Linux kernel is for me a kind of platform for running in low-level programming skills, I try to gradually improve my own work. The SysFS module for managing screen colors received minor syntax changes, however KCAL for Nexus 7 (Flo) has undergone significant changes: first, the panel color temperature is now changed using LUT registers. Unlike modern display controllers, our MDP4 controller does not provide a flexible interface for updating all IOCTL-accessible registers at one time, so the effect of altered values ​​of these registers can be manifested only when the panel image is updated simultaneously. LUT registers, on the other hand, cause the panel to be updated immediately, so the transition from the PCC to the LUT made it possible to make the color temperature change instantaneous. This, by the way, does not provide any other custom kernel. Color inversion works thanks to Inverse Gamma Correction registers; here, unfortunately, the behavior of the panel does not differ from that when the PCC registers are modified, so colors can be reversed only while updating the image on the panel.
  
  
• The interface used to change CPU-on-frequency voltages is rewritten into the framework format, i.e. It is universal in terms of the CPUFreq structure. Now it connects to the cpufreq driver, and does not go directly to the heart of the interface, which is unacceptable in Linux design. The acpuclk-krait interface itself received minor stylistic and syntactic edits. Nothing really interesting.
  
  
• At the same time, SysFS analogue of sound line gain control via IOCTL was rewritten in its entirety. This is a completely different interface that supports the registration of several hardware codecs at once and allows you to dynamically manipulate the gains of the required lines through SysFS. Also, like CPUFreq->VDD interface, my module is a so-called full interface, i.e. It is universal in terms of ALSA-compatible audio codecs. Intervention in the lines controlled through it by IOCTL is filtered using a simple filter in the function of recording to audio registers inside ALSA. This allows for fully dynamic register management, which implies the ability to take control of any audio register available in the hardware audio codec (see the header files wcd93xx_registers.h, where xx is the minor version of the hardware audio codec) and not worry about the conflict with IOCTL by customers.
  
  Map of the main sound registers hammerhead

  - Microphone in normal mode ->0x251 (TAIKO_A_CDC_TX7_VOL_CTL_GAIN)
  - microphone during video recording ->0x249 (TAIKO_A_CDC_TX6_VOL_CTL_GAIN)
  - Left channel headphone ->0x2b7 (TAIKO_A_CDC_RX1_VOL_CTL_B2_CTL)
  - Right channel headphone ->0x2bf (TAIKO_A_CDC_RX2_VOL_CTL_B2_CTL)
  - Left and right multimedia speaker channels (single channel sound) ->0x2e7 (TAIKO_A_CDC_RX7_VOL_CTL_B2_CTL)
  Map of the main sound registers flo

  - Microphone in normal mode ->0x249 (TABLA_A_CDC_TX7_VOL_CTL_GAIN)
  - microphone during video recording ->0x251 (TABLA_A_CDC_TX7_VOL_CTL_GAIN)
  - Left channel headphone ->0x2b7 (TABLA_A_CDC_RX1_VOL_CTL_B2_CTL)
  - Right channel headphone ->0x2bf (TABLA_A_CDC_RX2_VOL_CTL_B2_CTL)
  - Left channel multimedia speaker ->0x2c7 (TABLA_A_CDC_RX3_VOL_CTL_B2_CTL)
  - Right channel multimedia speaker (two-channel sound) ->0x2cf (TABLA_A_CDC_RX4_VOL_CTL_B2_CTL)
  
  
• The minimum and maximum driver brightness of the LM3630 (hammerhead) backlight can now be manually changed via the SysFS outputs / sys / class / backlight / lm3630 / device / lm3630_min [max] _level, respectively. This is a more correct analogue of Dimmer's favorite Backlight, used by many other cores. My core, in turn, allows you to edit the brightness thresholds used by the backlight driver directly. Accepted values ​​are in the range from 1 to 255 inclusive. You can see that by default the minimum backlight brightness on the Nexus 5 (hammerhead) is at level 5, i.e. you have the ability to make the minimum brightness significantly less than the one you are used to. However, LGE engineers didn’t just refuse to use the lowest possible brightness - the LM3630 at the hardware level is unable to provide uniform screen backlighting at a driver brightness below 3 (at least in the case of hammerhead). Take note of this.
  
  
• The kernel has the ability to switch between CD-ROM and USB emulation modes on the fly, using the sysfs outputcdromowned by the f_mass_storage platform driver. This is necessary for the correct operation of DriveDroid.
  
  
• The standard LED will now work forever if the off timeout (off_ms) is zero (logical, isn't it?). A similar method is used in the modified LightsHAL, used in many custom firmware to implement a notification about the status of the battery. Moreover, the support of a constantly active LED in the absence of a turn-off timeout can now be turned off. Judging by the custom firmwares branch, this is relevant for some Android Oreo users (there is no support in the kernel. It's funny, isn't it?). The corresponding SysFS outputs are in / sys / class / leds / red [green] [blue] / always_on. Yes, the always_on flag is set for each color separately. If you have a desire to disable support for the forever active LED completely, you can do it manually through the terminal / file manager or using the simplest Shell command:
  
  for i in red green blue; do echo 0>/ sys / class / leds / $ i / always_on; done
  
  
• Kernel Samepage Merging module has been added to the kernel. The module itself is strikingly different from that in the hammerhead drain kernel (Linux 3.4). The variant in my kernel almost completely corresponds to that in the official Nexus 5X kernel (bullhead) [Linux 3.10], not counting the activity during the system startup by default. I decided to leave KSM inactive, as it raises the overhead in the part of the virtual memory system responsible for the mapping of new pages. However, with current trends in terms of attitude to RAM, KSM can help reduce the workload of RAM and, as a result, improve multitasking. The choice between better multitasking and better performance is yours.
  
  
• The interface for controlling the voltages of the CPU as a whole does not differ from that in other cores (a variant from uvMyfluxi), however, unlike the others, it is implemented in a human way: without an incomprehensible conversion of functions into a completely different type (how did it work at all?), with protection in the form of a mutex and safe reading of the existing values ​​using the scnprintf function.
  
  
• IP (6) _NF_TARGET_TTL is active in the kernel. Those who need it know what it is. If this does not tell you anything, do not try to change the TTL value. This feature is extremely dangerous in that you can inadvertently create immortal network packets that will lead to the collapse of the network interface.
  
  
• The architectural timer now uses all 56 available architectural bits. Again, unlike the rest of the cores, I do not use the option of using all the bits from uv.Myfluxiwhich consists in porting the corresponding commit from u.Stephen Boyd, and which leads to random drops in the kernel due to the fact that the method of reading registers is not entirely correct on our architecture, but I use the register reading function already present in the kernel, simply removing zeroing of the upper 24 bits.
  
  
• I brought the frequency of the interrupt timer to the stock value - 100Hz. As it turned out, our old hardware is unable to work adequately with the triple number of interrupts, even despite the architecture of dynamic ticks. I noticed frequent cash misses and abnormally high overhead in some sections of the kernel, so let's leave it to newer generations.
  
  
• The entire architecture of nuclear timers, work queues and threads is reduced to the state of those in early Linux 3.8. It was necessary for me to get benefits from new internal calls, such as mod_delayed_timer, which made it possible to noticeably simplify some points. In addition, a more modern work scheduling architecture can reduce the cost of operations such as connecting / disconnecting logical cores, initializing and destroying work queues, reinitializing timers, etc. But the main goal was still to receive new calls. The rest somehow grew by itself. Honestly.
  
  
• The input event driver is now capable of handling several such events at once. At the same time, I saved the hardware calibration of monotonous time events. All this together made it possible to avoid extremely high intervals of being in a state of disconnected interrupts in moments of superstrong system load, coupled with a huge number of synchronous input events. This is quite useful on devices with increased interactivity, such as smartphones and tablets.
  
  
• Last time I forgot to report that the kernel is able to work with file systems such as NTFS and exFAT. Fans of connected storage must be satisfied.
  
  
• Finally, the main chip of the Breakfast, which is the central trump card and the basis of core competitiveness. Do your moods often drop after your favorite device suddenly freezes and reboots? After this depression eats you with your head? No problem! New Breakfast with its signature chip - a pulsating LED during power-up - will take you out of the abyss of despair. Beautiful iridescent LED charms, forcing to restart the smartphone more and more often. Spontaneous reboots are no longer annoying, on the contrary: they delight the eyes and soul. Unfortunately, Nexus 7 users are less fortunate: this tablet has a simple monochrome LED. But even so, a bright white light is better than nothing.

Download

| Version List -Github
| Special version for Android Oreo -

savagemessiahzine.com download

Additional Information

| Original post -savagemessiahzine.com
| Author's profile -savagemessiahzine.com
| Source -Github

Blu_spark

Blu_spark

Android 6.0.y - 7.x.y

Speklist

• Powered by Ubuntu 16.04.1 x86_64 (sparky VPS @ www.servercore.com.br - thanks @ChristianLetti)
  
• Build with custom toolchain blu_naro-7.2 (cortex-a15 neon-vfpv4 hard-float targeted, build with the latest Crosstool-NG from developer snapshot Linaro GCC 7.2-2017.10)
  
• Full-O3 support with device and target flags enhanced, linaro build improvements, etc
  
• Less stock is builds based on android-msm-hammerhead-3.4-marshmallow-mr3
  
• Kernel compressed with XZ, custom tweaked RAMdisk with GZIP (faster boot times)
  
• Removed some debug and logging options
  
• Enhanced performance and battery patches
  
• Several CPU Governors (blu_active own governor is default- fastlane mode available!)
  
• Voltage Control UV_mV_table (UV down to 700mV possible)
  
• Enhanced TCP methods (westwood is default), Network and Wifi tweaks
  
• Several I / O control (added FIOPS and ZEN is default), tweaked filesystems (F2FS, ExFAT, NTFS & CIFS available)
  
• Disabled stock mpdecision and thermald
  
• blu_plug: Dynamic hotplug for hammerhead
  
• msm_thermal simple driver (default 65 C threshold)
  
• Optimized RWSEM, AES and SHA1 routines (with neon support)
  
• Vibrator Strength tunable (60% default)
  
• KGSL fixes and reworked GPU driver (removed doubled freqs, goes to idle @ 27MHz), Simple GPU Governor by default
  
• Gamma Control and KCAL - Advanced color control for Qualcomm MDSS (RGB calibration and post-processing features)
  
• Backlight dimmer ready (disabled by default, tuned min manual brightness - level 3)
  
• USB automount (ROM must support it, otherwise use stickmount)
  
• CDROM emulation added to mass_storage (Drive Droid compatible)
  
• Kexec (multirom) full support
  
• doubletap2wake available (0: disabled, 1: enabled, 2: enabled fullscreen), option to disable with power button or after wake_timeout
  
• USB Fast Charge (USB mode up to 900mA with MTP), proprietary charger detection
  
• init.d support (put your scripts in /system/su.d), Kernel cleaner script
  
• Compatible with Kernel Adiutor, Universal Kernel Manager and others

Download

| Version List -XDA

Additional Information

| The official theme isXDA
| Author's profile -XDA
| Source -Github

Franco kernel

Franco kernel

Android 4.4.y - 8.1.0

Speklist

This core has no official speck list.

Download

| Version List -franco-lnx

Additional Information

| The official theme isXDA
| Author's profile -XDA
| Source -Github
| List of changes -Changelog
| Official app -Play Market

ElementalX

ElementalX

Android 4.4.y - 8.1.0

Speklist

• Easy installation with Aroma installer
  
• overclocking / underclocking options
  
• optional l2 and bus overclocking
  
• ElementalX CPU governor with gboost
  
• full voltage control
  
• choose voltage profile in installer
  
• wake gestures support
  
• optional sweep2wake and doubletap2wake
  
• options to disable wake functions with timeout or power button
  
• optional USB fastcharge
  
• NTFS r / w and exFAT support
  
• option to disable fsync
  
• multirom support
  
• optional backlight dimmer
  
• / dev / frandom support
  
• optional cooler color preset
  
• Advanced color and gamma control
  
• Sound control
  
• DriveDroid support

Download

| Official site -EX

Additional Information

| The official theme isXDA
| Author's profile -XDA
| Source -Github
| Official app -Play Market

Nitrogenextreme

Nitrogenextreme

Android 6.0.y - 7.x.y

Speklist

• Add Overclock CPU up to 2.5 GHz
  
• Add Underclock CPU up to 232 MHz
  
• Add CPU voltage control
  
• Add UKSM
  
• Add simple GPU governor
  
• Add msm hotplug
  
• Add msm turbo
  
• Add Use Android device as USB keyboard / mouse
  
• Add USB fast charging support
  
• Add faux sound control
  
• Add faux speaker control
  
• Add faux gamma control
  
• Add frandom
  
• Add fsync on / of
  
• Add ZRAM
  
• SeLinux permissive
  
• Add FIFO, SIO, SIO +, VR, ZEN I / O Schedulers
  
• Add abyssplugv2, lionheart, pegasusq, uberdemand, wheatley, hyper, darkness, nightmare, ondemandplus, optimax, smartmax, yankdemand CPU Governors
  
• Add stereo call recording support
  
• Add Wakelocks on / of
  
• Add kexec-hardboot
  
• Add af_unix: speedup / proc / net / unix
  
• Add backlight dimmer
  
• Implement always on RGB LED support
  
• Introduce builder and kernel packer
  
• Use ondemandplus default cpu governor
  
• Disable test iosched, uze VR by default
  
• Enable NTFS FS and cpu voltage
  
• init.d: ​​add config
  
• Enable LCD_KCAL
  
• Enable all tcp congestions
  
• Use westwood by default
  
• New installer
  
• XZ kernel compression
  
• Build with Linaro 4.9.4
  
• Overclock CPU up to 2.8 GHz
  
• block: scheduler: add ZEN v2
  
• Zen Sched: set fifo batch to 16 to reduce overload on eMMC and CPU
  
• usb: gadget: mass_storage: added entry sysfs for cdrom to LUNs
  
• Enable fastcharging by default
  
• Enable MSM hotplug by default
  
• Add franko thermal driver
  
• Add double tap to wake touchscreen driver
  
• Init included to ramdisk
  
• Powersuspend

Download

| Official site -xyyx-dev
| Stable versions -Github

Additional Information

| The official theme isXDA
| Author's profile -XDA
| Source -Github

Bricked kernel

Bricked kernel

Android 4.4.y - 6.0.y

Speklist

• Based upon Google's msm 3.4 source
  
• Various fixes, improvements and optimizatios (look @ github)
  
• Compiled with gcc4.7.2 toolchain (linaro 09.12)
  
• -O3 + optimized
  
• Snapdragon & CortexA15 optimizations
  
• replaced qcom's hotplug binary with msm_mpdecision (IN-KERNEL, better battery life + performance)
  
• Extensive sysfs interface for all users with tuneables you want (/ sys / kernel / msm_mpdecision /)
  
• replaced qcom's thermal binary with my IN-KERNEL solution. (/ sys / kernel / msm_thermal /)
  
• export krait version to: / sys / kernel / debug / krait_variant
  
• Allow OC up to 2.5Ghz
  
• Allow UC to 96Mhz
  
• Undervolting (faux123)
  
• F2FS Support
  
• Multirom support
  
• KCAL (savoca) & Gamma Control (faux123)
  
• Sound Control (faux123)
  
• Default clocks: 300 Mhz min & 2265,6 Mhz max

Download

| Official site -bricked.de

Additional Information

| The official theme isXDA
| Author's profile -XDA
| Source -Github
| Official app -Play Market

Hells-core

Hells-core

Android 4.4.y - 7.x.y

Speklist

This core has no official speck list.

Download

| Version List (Basketbuild) -Basketbuild
| Version List (Dropbox) -Dropbox

Additional Information

| The official theme isXDA
| Author's profile -G +
| Source -Github
| Official site -Blogspot

Uber kernel

Uber kernel

Android 4.4.y - 6.0.y

Speklist

This core has no official speck list.

Download

| Official site -uber

Additional Information

| The official theme isG +
| Author's profile -G +
| Source -Github

CAF core

Truesight

Truesight

Android CAF 5.x.y - 7.x.y

Speklist

• Built with LG PWR-CORE Toolchain (Linaro GCC 6.2.0)
  
• Optimizing toolchain flags: -O2, cortex-a15, neon-vfpv4
  
• Disabled more stuff of debugging
  
• Kernel compressed with LZ4 algorithm
  
• CPU govermors: interactive, conservative, ondemand (default), powersave, perfomance, impusle, alucard, darkness, nightmare, pegasusq
  
• IO sheduling: cfq, noop, deadline, row, fiops, zen (default)
  
• TCP congestion algorithm: cubic (default), bic, westwood, hstcp, htcp, hybla, vegas, veno, scalable, lp, yeah, illinois
  
• MSM_Hotplug (by @myfluxi).
  
• Thermal engine: 3-phase Bricked thermald (by @ show-p1984)
  
• F2FS support!
  
• New touch boost driver made by @ alucard_24
  
• MSM limiter driver v5.1 (by @dorimanx)
  
• CPU frequencies: 300-2.57 GHz (default 300-2.2 GHz)
  
• Modified cpufreq driver, you can set min / max freq and governor on each core!
  
• Ported cpuidle driver from 3.10
  
• GPU frequencies: 27-650 MHz (27 idle, 100-650 MHz work)
  
• New Adreno GPU Snapshot driver from LA.DR.2.2.0_rb1.00
  
• Sound control: FauxSound (by @ faux123)
  
• Power Efficient Workqueues
  
• Gentle Fair Sleeper (Disabled on default)
  
• ArchPower (Disabled on default)
  
• USB FastCharge (Disabled on default)
  
• Fsync ON / OFF trigger
  
• Async fsync
  
• State notifier driver made by @ neobuddy89
  
• Adreno idling algorithm v1.1 (by @ arter97)
  
• Simple gpu algorithm (by @ faux123)
  
• ZRAM support (disable by default)
  
• ZSMALLOC
  
• UKSM (disable by default)
  
• Voltage Control v1.2 (by @dorimanx) Warning! High undervolting may your phone calls freezes and reboot.
  
• Gamma control
  
• MultiROM support (kexec-hardboot)
  
• Backlight Dimmer (disable by default)

Download

| Version List -AndroidFileHost

Additional Information

| The official theme isXDA
| Post onsavagemessiahzine.com - savagemessiahzine.com
| Author's profile -XDAandsavagemessiahzine.com
| Source -Github

Irrelevant / Unpopular kernels

Android marshmallow

• Big bum(big_bum)AOSP
  XDA thread | Download
  
  
• Despair kernel(DespairFactor)AOSP
  XDA thread
  
  
• Racer kernel(rachanta)AOSP/CAF
  XDA thread

Android Lollipop

• Chaos kernel(neobuddy89)CAF
  XDA thread | Download | Mod for 12.1
  
  
• Bluemoon(Matthew_333)CAF
  XDA thread | Download
  
  
• Big bum(big_bum)AOSP
  XDA thread | Download
  
  
• Despair kernel(DespairFactor)AOSP
  XDA thread
  
  
• Maximus(nexterultim)AOSP
  Download | Old version
  
  
• Brain explosion(garynych2010)AOSP/CAF
  Download | FAQ
  
  
• Hybrid(neobuddy89)AOSP
  XDA thread
  
  
• Faux kernel(faux123)AOSP
  XDA thread
  
  
• Racer kernel(rachanta)AOSP/CAF
  XDA thread
  
  
• Phenom kernel(TomorrowLand_Ace, former Simplicist, CharizardX)AOSP
  XDA thread | Download
  
  
• Chroma.Kernel(TomorrowLand_Ace)AOSP
  XDA thread | Download
  
  
• Electroactive Kernel(Electrex)AOSP
  XDA thread | Download

Android Kitkat

• Linaro(LaboDJ)AOSP
  XDA thread | Download, Mirror
  
  
• Franco kernel[Phantom] (PhantomGamers)AOSP
  XDA thread
  
  
• Hybrid(neobuddy89)AOSP
  XDA thread | Download | List of changes
  -> recommended settings (new)
  -> recommended settings
  
  
• Vanir(styr)AOSP
  Download
  
  
• Arkenstone(Gnome)AOSP
  XDA thread | Download
  
  
• Brain explosion(garynych2010)AOSP
  Download
  
  
• Faux kernel(faux123)AOSP
  XDA thread | Download
  
  
• Simplicist, former CharizardX(TomorrowLand_Ace)AOSP/CAF
  XDA thread | Download | List of changes
  
  
• Big bum(big_bum)AOSP/CAF
  XDA thread | Download
  
  
• Chaos(neobuddy89)AOSP/CAF
  XDA thread | Download | List of changes
  
  
• Furnace(Savoca)AOSP/CAF
  XDA thread | Download
  
  
• Ultimate / Alternative(mcaserg)AOSP
  Download
  -> tuning
  
  
• [DER KERNEL] Trinity(morfic)AOSP
  XDA thread
  
  
• Fugodod(renaud)AOSP
  XDA thread
  
  
• n3oHammer(n3ocort3x)AOSP/CAF
  XDA thread
  -> battery saving settings
  -> performance tuning
  
  
• TeamAT Next(AuxXxilium)CAF
  XDA thread
  
  
• Next saber(TomorrowLand_Ace)CAF
  XDA thread
  
  
• Linear(hottix)AOSP
  XDA thread
  
  
• Dr. West's Re-Animated(koe1974)AOSP
  XDA thread
  
  
• Mahdi Experimental(Mahdi)AOSP
  XDA thread

Kernel core

Android marshmallow

• Stock core Android 6.0.1 M4B30Z
  
• Stock kernel Android 6.0.1 M4B30X
  
• Stock kernel Android 6.0.1 MOB31E
  
• Stock kernel Android 6.0.1 MOB30Y
  
• Stock kernel Android 6.0.1 MOB30P
  
• Stock kernel Android 6.0.1 MOB30M
  
• Stock kernel Android 6.0.1 MOB30H
  
• Stock kernel Android 6.0.1 MOB30D
  
• Stock kernel Android 6.0.1 MMB29X
  
• Stock core Android 6.0.1 MMB29V
  
• Stock kernel Android 6.0.1 MMB29Q
  
• Stock kernel Android 6.0.1 MMB29S
  
• Stock kernel Android 6.0.1 MMB29K
  
• Stock kernel Android 6.0.0 MRA58N
  
• Stock core Android 6.0.0 MRA48K

Android Lollipop

• Stock core Android 5.1.1 LMY48M
  
• Stock kernel Android 5.1.1 LMY48I
  
• Stock core Android 5.1.1 LMY48B
  
• Stock core Android 5.1.0 LMY47I
  
• Stock core Android 5.1.0 LMY47D
  
• Stock kernel Android 5.0.1 LRX22C
  
• Stock kernel Android 5.0.0 LRX21O

Android Kitkat

• Stock core Android 4.4.4 KTU84Q
  
• Stock core Android 4.4.4 KTU84P
  
• Stock core Android 4.4.3 KTU84M
  
• Stock core Android 4.4.2 KOT49H
  
• Stock core Android 4.4.0 KRT16M

Survey Screenshots

2015 survey

Survey for 2016

Cap updates

From 02.17.2018:

• Update information.
  

From 09/25/2017:

• Added specklist kernel Breakfast.
  

From 09/22/2017:

• Removed spoiler"Useful links", completely lost relevance a year ago.
  
• Minor stylistic changes for better readability.
  

From 09/16/2017:

• Easy update caps.
  

05.06.2017:

• Added core Breakfast. Again.
  

From 11.05.2017:

• Removed the core of Breakfast.
  

From 03/20/2017:

• Added core Breakfast.
  

From 02.21.2017:

• Improving the layout of the caps for the correct display in the mobile client.
  

From 04.02.2017:

• Minor fixes and improvements.
  

From 01/04/2017:

• Improved layout and convenience of the caps.
  

From 12.15.2016:

• Added stock kernel Android 6.0.1 M4B30Z.
  

From 11/30/2016:

• Final redesign of the caps.
  

From 10.27.2016:

• General redesign caps.
  

From 10/13/2016:

• Added coreBlu_Spark Modifiedin "Kernel for 7.0".
  

From 09/25/2016:

• Added coreElementalXin "Kernel for 7.0".
  

From 09/24/2016:

• Added coreNitrogen EXin "Kernel for 7.0".
  

From 09/21/2016:

• Added coreFrancoin "Kernel for 7.0".
  

From 09/14/2016:

• Added coreStocklikein "Kernel for 7.0".
  

From 09/05/2016:

• Added item"7.0"in modified nuclei.
  
• General fixes caps.
  

From 07.16.2016:

• Addedfix WiFiin spoiler "Important Information".
  
• There is also a link tonew GPU drivers.
  

From 05/31/2016:

• Added coreMaximus.
  

From 05/30/2016:

• Some fixes caps. Thank uv.Vitromanof.
  

May 24, 2016:

• Added new settings from uv.Vitromanof - Flawless.
  

04/05/2016:

• Added coreStocklike.
  

From 09.03.2016:

• The cap is edited in accordance with the new rules from 09.03.2016.
  

03/08/2016:

• Added new settings for the governor from uv.Vitromanof.
  

03/05/2016:

• Added coreBlu_Spark Modified.
  

03/04/2016:

• Added settings of the governor from uv.Vitromanof.
  

02.02.2016:

• Added coreTruesightfrom uvdemon1992.
  

From 01/09/2016:

• Added programs to configure the kernel.
  
• Updated rules.
  

07.01.2016:

• Redesign caps

The curator of the topic isneXTerUltim (QMS)

Post has been editedneXTerUltim - 19.07.18, 14:56

Reason for editing: Updating information

[Anshe2]

Here I sit in this topic for two days already. 40 people installed custom kernels, but almost no one unsubscribes how the cores behave (at least to compare with the drain), speed (although n5 flies on the drain), energy saving, maybe which chips appeared in the core. At least briefly, that the owners of n5 are approximately oriented to whom what is needed and for what conditions of operation of the animal. Yesterday I installed ElementalK, I disconnected it from the socket at 23.00 and did not touch it at all until 9.00 in the morning. The result of minus 25% is not good (WI-FI, 3G, location search is disabled), I have eaten 2-4% on SGS3 during this time. Who has this core, what are your results?

[Able373997]

Anshe2,
I almost did not sit on the stock, so I can not say anything. And all the time on ElementalX. Now I will put 0.11 version. It's strange of course why you ate so much without using services. It just happened, tel. lay the night and the service off, ate a few percent. And usually I charge for charging from a computer for the night. This morning, for almost 1.5 hours, the computer was tormented with its 20%. Google + and youtube from it looked through. I kept the charge very well, although the brightness was at a minimum. To quickly kill Akum, I decided to start the game. then a reliable way: D I’ll check how I’ll keep it at night, accomplish your goal.

And I hope you do not have the 2nd touch and scrolling to wake up. These features theoretically put the battery 20% per night

[Anshe2]

Able373997,
Scrolling on waking up is on him, and I thought that he would not let the smart fall asleep, but what is the 2nd touch and where to look for it?

[Able373997]

Anshe2,
Well, this is when the core was set, there was a choice for you to scroll ... and the 2nd touch (as on the LG G2, the phone is unlocked when you double tap on the screen). You should reflash the kernel again and, where they ask which features to choose from this, choose nothing. and will sleep normally tel.

[Skazki]

Able373997,
what for?https://play.google.com/store/apps/details?...ermod&hl=enIs there a disconnect for this elemental?

[XetreX]

Able373997 @ 12.12.2013, 11:43

Anshe2,
Well, this is when the core was set, there was a choice for you to scroll ... and the 2nd touch (as on the LG G2, the phone is unlocked when you double tap on the screen). You should reflash the kernel again and, where they ask which features to choose from this, choose nothing. and will sleep normally tel.

Eats 1-2% of the battery per hour (because it constantly keeps the touch in standby mode), the chip is interesting, but IMHO it's not worth it

[Able373997]

ElementalX 0.10 for 7 hours ate 9% of Akum. This = votes 15min call + 25min screen + Google service 5min + A little game (Internet included)

[email protected], 11:47

what for?https://play.google.com/store/apps/details?...ermod&hl=enIs there a disconnect for this elemental?

Yes you are right, there are such parameters here. Thanks for the tip!

[Anshe2]

Reinstalled the core ElementalX 0.11 already without any scrolling and double tapes, for the night (7 hours) it took 3%. I am glad and happy! I also wanted to know the opinion of someone who has a free core, how their results are.

[ado1]

I tried to install franco via adb, but the waiting device has been hanging for half an hour. Adb sees the phone, he is in recovery. what to do?

[M.A.R.K.]

I put the core of the franco through his prog ... duck the phone began to noticeably lag especially in games, there were strong FPS subsidence!
and the maximum frequency was only 1.9 MHz for each core ... I put back a happy core from the backup and for some reason I lost the standard limit on the processor frequencies, now even at 65 degrees of heating the percent runs at 2.3 MHz
I did not understand what happened :))

[Able373997]

[email protected], 07:03

Reinstalled the core ElementalX 0.11 already without any scrolling and double tapes, for the night (7 hours) it took 3%. I am glad and happy! I also wanted to know the opinion of someone who has a free core, how their results are.

I ElementalX 0.11 per night 1% ate. turned off all services.

M.A.R.K. @ 12/02/2013, 17:47

now even at 65 degrees of heating the percent runs at 2.3 MHz

Probably put a stock core core

[flatline]

ado1 @ 12/02/2013, 15:31

tried to put franco via adb

I installed it for the first time via ADB - well, exactly exactly as it is written in the header on the kernel on the XDA - fastboot flash boot boot_name_img. Is that the trick is that the phone must be on the ADB is connected, and not MTP. Then I installed CWM Recovery - well, everything is fine here - I selected install zip in recovery, I chose the kernel and everything was done in a couple of seconds.

As for the franco, I myself can’t say anything yet, because core just a day and a half. Installed on top of Cataclysm ROM, a rather popular Spark on XDA. In a couple of days I will write off what and how the core.

[NikitaNikolaevich]

It is the core of Franco's noramal flight. Reduced the frequency to 1.4GHz. Without lags and friezes. It is also worth the software from qualcom BattaryGuru can be taken in the play market. While pleased with the core and power consumption.

[garynych2010]

Able373997 @ 12.12.2013, 23:53

I ElementalX 0.11 per night 1% ate. turned off all services.

coren3oHammer
the same is perfectly asleep, by the way, I don’t have the habit of disconnecting anything .... for the night 1-2% eat and it works great in general

Post has been editedgarynych2010 - 03.12.13, 11:54

[vasya_krets]

My question is: how does a custom kernel (like ElementalX) behave in the ART environment?

[Able373997]

Vasya_krets @ 12/03/2013, 15:11

My question is: how does a custom kernel (like ElementalX) behave in the ART environment?

I also have ElementalX 0.11 + ART + Read Ahead 2048. This is what it does:
This is 77 minutes of conversation.
Screen 160 minutes
Wi-Fi 10 hours
Phone hours 47 hours

Post has been editedAble373997 - 03.12.13, 20:52

[Anshe2]

Able373997,
Read Ahead 2048. What is it?

[arbi97]

Anshe2,
Clipboard!

[garynych2010]

Anshe2 @ 12/03/2013, 22:35

Read Ahead 2048

Arbi97 @ 12/03/2013, 10:39

Clipboard

Able373997 @ 12/03/2013, 21:36

Read Ahead 2048. That's what makes

and with me, dancing with this, on 4.4, is only carried out to deterioration ... that is, if on the old smart with 4.1.2 I stood at 2048 and the reading speed grew from 27 to 72
judged by measurementssd tools
and here (on 4.4), firstly, the prog does not plow, and secondly, the value is above 256, leading only to tormas (judging by the decision of the antitu)
so personally, while I left 256 (just Western coremakers, usually for our device, it is the current value that they shove by default in their cores, judging by the source codes of the cores)

Post has been editedgarynych2010 - 03.12.13, 22:21

[garynych2010]

patriotman @ 12/04/2013, 07:56

I set up and here I was puzzled by the choice of IO of the sheduler :) I saw a new for myself and an unfamiliar ROW% more precisely)
This is how I understand

comparison
I / O Shedulers:

I use or ROW, as the most responsive, or ZEN, as the most productive and suitable for daily use.

Noop
This scheduler combines all received requests, it is best used together with flash drives, because unlike a hard disk, a flash disk does not require the redistribution of several requests due to the absence of mechanical moving parts.
Pros:
+ Uses a minimum processor, paralleling to kernels
+ Best for flash drives
+ Good database bandwidth
Minuses:
- The smaller the cores, the more the performance drops.
- With increased processor load, it sags significantly in performance

Deadline
The goal is to reduce I / O delays and reduce the response time to a request. Five queues aggressively change the order of incoming requests.
Pros:
+ Close to real time
+ Small I / O delays
+ Best for accessing databases and best for querying
+ Like noop good for flash drives
Minuses:
- When the system is loaded, it can hellishly lag.
- with an extremely high load, some processes will be cut off by a kind of “deadline” (hence the name) - you may sometimes have encountered a strange phone behavior: you set 30+ programs to update, but only some of them were updated, and the rest didn’t seem to start.

CFQ
Well, CFQ for I / O planners is akin to Ondemand for Governors. It is the most balanced and focused on performance in most cases. However, in android some things work differently, so this is not the most suitable I / O scheduler. The fairest planner, trying to allocate resources equally
Pros:
+ Most balanced by I / O
+ Just to set up
+ Is allocated on multi-core systems
+ Better performance with Databases after Deadline
Minuses:
- Since resources are divided equally, then relatively one task is slow.
- Sometimes the delay may be too big (jitter)

Bfq
Based on CFQ, inferior in VR and SIO performance, even if it is a version of BFQ v2. (having the highest speed in USB data transfer).
Pros:
+ Good USB transfer rate
+ Considered best for recording heavy video or streaming (Jitter is smaller than CFQ)
+ Accurate I / O Planner
Minuses:
- Bad for benchmarks

SIO (Simple IO)
As the name implies, Simple I / O is one of the simplest. The simpler the I / O schedulers for android devices, the better - this is exactly the case. It is reliable, not as good as the VR at the maximum, but nevertheless one of the best in performance. But at times, performance can become very draining, friezes.
Pros:
+ Reliable
+ Minimum delay for requests
Minuses:
- Slow random read speed on flash drives compared to other schedulers
- Sequential read speed is also not up to par

VR
It is considered the most productive, in principle it is. Its performance is not constant, it can show both maximum and minimum. But when he shows the maximum - he is the best. But he is not very stable. It has some elements of Deadline.
Pros:
+ Best for benchmarks, as its peak performance is at peak load.
Minuses:
- Unstable, unreliable
- Not suitable for daily use

ROW
Read over Write - that is, the priority goes to read, and then write. It was created for better responsiveness, based on the fact that on mobile devices we will not have as many parallel threads as on computers. Usually it is 1 stream or, at best, 2.
Pros:
+ Good speed for reading data (viewing photos, reading from db, etc.)
+ If there are read requests, then execute them as soon as possible.
Minuses:
- Poor data writing performance (during data transfer, installation of applications, etc.)
- Requests for recording accumulate for a long time in the queue

Zen
In general, it is similar to VR and SIO, but it is simpler and closer to Noop, it also works, but it differs slightly in work (this is a translation from clumsy English). Zen does not sort anything and serves in the order of turn - first come, first served (completed). Synchronous requests from it are more priority than asynchronous.
Pros:
+ Requests for reading will never wait
+ Like Noop good performance for reading on flash drives
Minuses:
- Slow recording speed even on fast flash drives, but judging by the tests, it is not.

diagrams (first already from kernel 3.4)

But if you pass through an online translator, then what is written about row in the kernel source

ROW I / O scheduler gives priority to READ requests over write requests when dispatched, without extra write requests. Requests are stored in the priority queue. Dispatching is done in an RR manner, when sending a quantum for each queue, is calculated in order of priority. Most suitable for mobile devices.

On an old smart, I am in a row row, here it is something IMNO is not ice, so while I stopped at cfq

Post has been editedgarynych2010 - 04.12.13, 11:55