[KERNEL] MiRaGe - for Nexus 4 stock ROM 10/25/15

[mrg666]

MiRaGe is a lean and efficient kernel for the stock Nexus 4 ROM with the optimizations and updates that are not included in Google's stock kernel. MiRaGe kernel fits squarely in the stock Nexus 4 ROM; all of the modules are integrated in the kernel just like the stock kernel and it should work with all AOSP ROMs that work with the stock kernel and boot image. However, only the current stock ROM is tested. If you decide to use MiRaGe, just flash and forget it since I have avoided adding more sysfs parameters. It is not my goal to enable all possible tweaking options and add every possible feature to the kernel such as multitude of governors, io schedulers, sweep2wake, fastcharge, etc. This kernel is not intended as a tweaker's kernel. You can, of course, tweak it as much as you want since that is your phone and kernel. But please try removing your tweaks before posting any problems. I always test the latest builds with the current stock ROM before posting here.

I am sharing exactly what I have developed for myself and posting here so that I can return at least a small part of what I have received from the open source community. I thought the amount of time I have spent for MiRaGe could be useful for others as well. In short, take it if you want it, leave it if you don't. But comments, suggestions are always welcome when they make sense.

Source Code:
Source code is based on Google's msm kernel source (currently android-msm-mako-3.4-lollipop-mr1.1) and a summary of my changes are below. You can find the full details of my changes and the complete source code in my repo.

Changes:
- synced with mainline Linux 3.4.110
- cleaned up kernel configuration and removed many unnecessary options
- removed kernel debugging options
- built with the Linaro toolchain (gcc 4.9.4 - 15.06) using standard krait and -O2 optimization
- removed AOUT and OABI support
- disabled both user-space msm_mpdecision and kernel-space msm_mpdecision
- removed msm_run_queue_stats, dcvs, and stock msm_mpdecision in the kernel
- added autosmp, a simple and efficient (by me) multi-core cpu hotplug driver
- disabled the user-space thermald and switched to kernel-based msm_thermal
- replaced CFQ with the latest BFQ as the default IO scheduler
- backported random and prandom updates from Linux 3.13 (no entropy depletion anymore)
- backported workqueue from Linux 3.8 to include many important improvements
- backported rwsem from Linux 3.11 to include lock stealing improvements
- backported mutex and rcu locking from Linux 3.10 and 3.8, respectively
- backported slub memory allocator updates from Linux 3.8
- backported cpufreq driver, ondemand, and conservative governors from Linux 3.12
- updated interactive CPU governor from AOSP and CAF
- disabled userspace CPU governor,
- enabled callback-free CPUs (RCU_NOCB_CPU)
- backported TCP Small Queues and CODEL net scheduler from Linux mainline and set as default
- updated kernel scheduler, msm-hotplug, msm-idle, msm-pm code from CAF and Linux mainline
- applied patch [v4] binfmt_elf.c: use get_random_int() to help with entropy depleting
- enabled autogroup scheduler and applied patch per-uid task group for Android
- added optimized ARM RWSEM algorithm
- added optimized ARM SHA1/AES routines
- enabled CPU-supported unaligned accesses
- disabled gentle fair sleepers in scheduler
- updated Qualcomm HW RNG driver from CAF
- enabled BPF JIT compiler for packet filters
- applied glibc patch to improve the performance of memcpy and memmove
- applied word-at-a-time ARM API patches
- enabled CPU overclocking up to 1.728 GHz with user-space vdd control
- optimized vdd curves, L2 and bus speeds for better performance and efficiency
- removed unneeded a2xx and a4xx components from kgsl driver
- modified the prima wifi driver to disable debug code
- removed PMEM completely, MiRaGe is pure ION
- add support for kernel mode NEON and NEON acceleration
- add NEON optimized SHA1, SHA256, and SHA512 crypto code
- add LoUIS API for cache maintenance ops to improve cpu hotplug latency
- added and enabled power_efficient workqueue
- added and enabled msm memutils
- added screen gamma, user space cpu voltage control, and dt2w
- backported devfreq driver from CAF and switched kgsl 3d governor to simple_ondemand
- backported many other fixes/updates/optimizations from CAF and Linux mainline, see the repo for details
- init.d supported if /etc/init.d and busybox are available
- a diff file of changes to ramdisk is here

Downloads:
Boot image for stock ROM:
LMY standard kernel Built: 10/25/15 MD5sum: a315cc446499d60cb4b3a61ea7bfa8f8
LMY overclock kernel Built: 10/25/15 MD5sum: 7c72a66830f511b025968db2bb743429

Anykernel updater for custom ROM:
Revert back to stock kernel to restore the original ramdisk and flash anykernel package of MiRaGe after that. This is not needed in the next anykernel update.
LMY standard kernel Built: 10/25/15 MD5sum: 85b4136ac0ada793da7b80763193095a
LMY overclock kernel Built: 10/25/15 MD5sum: 2924fb6a963b40087c296a7b1abfc1d3

KTU standard kernel Built: 10/31/14 MD5sum: dca7d06933eb43c8da3ba7941bb6ac88
KTU overclock kernel Built: 10/31/14 MD5sum: dca7d06933eb43c8da3ba7941bb6ac88

The only difference between the standard and overclock builds is the ~100mV undervolt in the overclock build. Both kernels have maximum CPU_freq = 1.728 GHz, default CPU_freq = 1.512 GHz, overclocking, and user space cpu voltage control enabled. Since the CPU gets hot quickly in Nexus 4, I only recommend overclocking with the overclock build that has built-in undervolt. If the phone doesn't boot with overclock kernel, it means that your CPU is not able to handle the undervolt settings. In that case, you can just reboot into recovery and flash the standard kernel. No-frills CPU Control is recommended to set the max overclock frequency. Each CPU has different overclock/undervolt ability. Don't get disappointed if the OC build doesn't work for you.

Installation:
You can do one of the followings
- Flash the zip files in recovery, there is no need to wipe cache or dalvik-cache
- Flash the boot image in the zip file using either Flash Image GUI or fastboot
- Here is the original boot image for LMY48I build, in case needed for going back to stock. Either flash in the recovery or open the zip file to extract the boot image.

Credits:
- Special thanks to Linux, Google, CAF, Linaro developers in general.
- @tvall, @bedalus, @xboxfanj, @ihancioglu, @xenyz for collaboration
- @stratosk for the screen gamma interface and dt2w
- @defconoi for collaboration (see Unleashed Kernel Series)
- @mathkid95 for the any-kernel updater package
- @joeykrim for FlashImageGUI
- @Christopher83 for the optimized Linaro toolchain builds
- Other credits are given in the repo for each commit

 

[mrg666]

Recommendations:

• I am frequently receiving requests to add sound patches in the kernel. I agree that the sound is not very good but there are solutions. I am using the Viper4AndroidFX as a replacement sound processor. I recommend giving it a try. You need to go to the sound options and select ViPER4AndroidFX to use this sound processor or freeze MusicFX (I use Link2SD for this). There is plenty of information at the above link. With this available, I am not planning to add any sound patches.

• Another frequent question is about choosing CPU governor and IO scheduler. In the earlier builds, interactive governor had the best balance of performance and battery life among other CPU governors and it is still available. In the latest builds, ondemand governor was backported from Linux 3.12 and replaced interactive as the default. The latest patches in the mainline Linux, especially stratosk's patch that optimized the load calculations made the new ondemand governor the better option regarding both power and performance. Regarding IO scheduler, BFQ scheduler has the best overall real-use performance and it is actively maintained/improved. You can use Nofrills CPU Control to change the governor and scheduler. But I would leave the defaults as BFQ scheduler and ondemand governor.

• Since all of the cpu power control functions are contained in the kernel with MiRaGe, the userspace PowerHAL library will be giving the following messages in the logcat.
  

  E/PowerHAL( 511): touch_boost: failed to send: No such file or directory
  E/PowerHAL( 511): touch_boost: failed to send: No such file or directory
  E/PowerHAL( 511): touch_boost: failed to send: No such file or directory

  These are harmless but if you want to eliminate them, just make a backup and delete/rename /system/lib/hw/power.msm8960.so and power.mako.so. The single purpose of touch_boost is to enhance the system response to the user interaction. But using a service in the user space to send a touch boost signal to the kernel via slow sysfs file system is the wrong way of trying to achieve lower latency. In addition, every touch input doesn't need a CPU frequency boost which wastes battery power. The best way of achieving the low-latency system response to user interaction is improving the efficiency of existing CPU governor which raises the CPU frequency and hotplug driver which enables off-line cpu cores when needed. In MiRaGe, CPU freq is only controlled by the CPU governor based solely on the CPU load and the latency is low since efficiency is improved by reducing such unnecessary bloat. Additionally, highly-efficient autosmp hotplug driver works in-sync with the CPU governor to enable off-line cpu cores when the the CPU frequency reaches a high threshold and still more compute power is needed. Therefore, touch boost bloat is removed.

• With some of the custom ROMs, root is lost after flashing MiRaGe because of using the init scripts in the ramdisk for starting the su daemon. SuperSU is the recommended solution. I might switch to any-kernel-updater to address this problem but as written in the OP, MiRaGe is primarily for the stock ROM. Also, having the full boot image in the zip file is more reliable than expanding/processing/repacking the boot image.

• MiRaGe supports init.d if it is setup. To setup init.d do the followings either within ES File Explorer or terminal .
  - install busybox (I use busybox on rails)
  - create /system/etc/init.d and chmod to 755 (rwxr-xr-x)
  - create your init scripts in the /system/etc/init.d directory. Name them 01yourscriptname (e.g. 01mysettings) and chmod 755. Make sure they are UNIX format (not in DOS/Windows).
  example:
  

  #!/system/bin/sh
  echo 1 > /sys/devices/virtual/input/lge_touch/dt_wake_enabled

  - reboot

• Here is how to add multiROM support

How to build:
If you are going to distribute your builds, please don't build your binaries with the same name (i.e. MiRaGe) and distribute in this thread. I would recommend you to start an alternative thread. Otherwise the problem reports will be too confusing for everyone.

First requirement is an ARM toolchain for cross compiling, i.e. using an X86 computer to generate ARM binary. I use Linaro tool chain for cross compiling like many others since Linaro specifically develops tool chains that produce optimized binary for ARM architecture.

Linaro toolchains can be downloaded from Linaro binary page. Christopher83 has built the latest Linaro-14.08 toolchain based on gcc-4.8.4 which is stable/reliable and I recommend starting the development with this toolchain.

The binary Linaro toolchain for Linux package needs to be expanded in a certain directory, probably inside the home directory. The source code for kernel is available in my Github repo, You can either download the kernel source as a compressed package or you can git-clone it with the following command (you will need git installed in your Linux computer)

git clone https://github.com/mrg666/android_kernel_mako.git

The kernel source can again be in a specific home directory.

After the source and toolchain are prepared, copy the configuration file for shooter, arch/arm/configs/mako_config, as .config to the root of the kernel source and use the following command to build the kernel

make ARCH=arm CROSS_COMPILE=~/untarred-toolchain-dir/bin/arm-linux-gnueabihf- zImage -j8

Replace j8 in the above command according to the number of cpus you have on your computer.
Also set CROSS_COMPILE based on the directory you have expanded the binary toolchain package in your home directory.

I always use the latest version of Xubuntu x64 (with custom built kernel) on my Linux workstation that has a AMD FX-8320 (overclocked to 4.2 GHz), 8 GB RAM, 500 GB HD. The compile time is about 2 minutes for me using all 8 cores. I have been using Ubuntu since version 10.04 to build Gingerbread, Jellybean, and Linux kernel and updated the OS to each and every new version, all of them worked just fine. There is no magic version of Ubuntu. The build problems arise from the package requirements not the OS version.

The flash package is easy. Just use any-kernel updater package in the OP as a template and replace zImage in /kernel directory with your build. If you want to create a boot image, see this post

Now that you have source and can build the kernel, you can add all the features you want to your own kernel

 

[hasanandnexus]

Woww greatt, thanks mirage

 

[Maelstr0m]

many thanks Mirage!
Does JSS come with caf video driver or it can be flashed on non-cm roms without problems?

Inviato dal mio Nexus 4 con Tapatalk

 

[alen1901]

Good to see new kernel which goal is simplicity, not many of them are here. ill try it when clean instal comes to repertoar. just one question, you didnt mention -O3 and gcc 4.8.2, so i asume you didnt use them? Thanks.

Poslano sa mog Nexus 4 koristeći Tapatalk

 

[mrg666]

I haven't updated the video driver from CAF ... yet. I will do after the 4.4 update if Google hasn't done yet.

Kernel is compiled with gcc 4.7.4 using -O2 optimization. gcc 4.8 was not giving me reliable builds yet. I will switch when 4.8 becomes stable. I have tried O3 optimization in the past and I didn't see any benefit of it. Plus, O3 optimization caused reliability issues especially with the latest gcc compilers.

MiRaGe should be compatible all AOSP-based ROMs, as long as the same user-space libraries are used with the stock 4.3 ROM. I can't claim universal compatibility since even stock JSS and JWR builds need different kernels.

 

[C.T.Richter]

Would you consider making a ZIP?
I found your kernel to be quite interesting, but I don't really like flashing via IMG file.

 

[ak]

Would you consider making a ZIP?
I found your kernel to be quite interesting, but I don't really like flashing via IMG file.

you can download one of 1000 kernels around here and replace the kernel.img ... and wholà you have a zip version.

 

[playya]

you can download one of 1000 kernels around here and replace the kernel.img ... and wholà you have a zip version.

Can't agree with you more but the problem with so many people on xda is that they rarely even open up a zip file to see how it operates. Guarantee most of the people will shy away simply because of the lack of a zip. Again I agree its not that hard to do but lets be honest here most have problems searching so taking a boot.img and putting it in a zip probably aint happening

 

[mrg666]

I have just uploaded the alternative flashable zip files. I will remove the image files since they are in the zip files now.

 

[QUIETLYloud]

Just flashed on slim and lost root... Just a heads up

Sent from my Nexus 4 using XDA Premium 4 mobile app

 

[mrg666]

you can download one of 1000 kernels around here

LOL that's right!

 

[mrg666]

Just flashed on slim and lost root... Just a heads up

Sent from my Nexus 4 using XDA Premium 4 mobile app

That can't happen due to flashing the zip files in the OP since there is nothing in the zip file that touches /system, it is not even mounted.

 

[DontPushButtons]

Gonna give it a go on Vanir. I'll report back of my root gets effected. Happened a lot with 4.3 when it first came out

Sent from my Nexus 4 using Tapatalk

 

[mrg666]

Gonna give it a go on Vanir. I'll report back of my root gets effected. Happened a lot with 4.3 when it first came out

Sent from my Nexus 4 using Tapatalk

I am using CWM 6.0.4.4. It is constantly suggesting to restore my root although it is not lost. If this is what is mentioned here, just ignore it; root is not lost. SuperSU, su keep working. Actually, I am losing my patience with CWMT lately. Maybe it is time to switch to TWRP.

 

[DontPushButtons]

I am using CWM 6.0.4.4. It is constantly suggesting to restore my root although it is not lost. If this is what is mentioned here, just ignore it. Root is not lost. SuperSU, su keeps working. Actually, I am loosing my patience with CWMT lately. Maybe it is time to switch to TWRP.

I'd say it's long overdue to switch to twrp lol. Ever since I switched to twrp back on my rezound, I have NEVER looked back to cwm. Not to say cwm isnt/wasn't great.. But you know how it is lol.

Sent from my Nexus 4 using Tapatalk

 

[mrg666]

I'd say it's long overdue to switch to twrp lol. Ever since I switched to twrp back on my rezound, I have NEVER looked back to cwm. Not to say cwm isnt/wasn't great.. But you know how it is lol.

Sent from my Nexus 4 using Tapatalk

I just don't like the "overdesigned" interface of TWRP. It is too fancy for my taste. But as long as it works when needed, it would be fine with me. I don't boot into recovery so frequently anyway since I flash the kernel in fastboot or Flash Image GUI.

 

[C.T.Richter]

Oddly enough, I lost root. I'm currently running the latest version of Paranoid Saberdroid.

 

[mrg666]

Oddly enough, I lost root. I'm currently running the latest version of Paranoid Saberdroid.

Is it just what the recovery says? I use the stock JWR ROM and root is preserved when I flash although CWMT falsely complains about it.

Edit: I just tested with TWRP as well. Root is still preserved.

 

[C.T.Richter]

Is it just what the recovery says? I use the stock JWR ROM and root is preserved when I flash although CWMT falsely complains about it.

Edit: I just tested with TWRP as well. Root is still preserved.

Using TWRP there is no error message, it just says it was installed successfully.
I just tried it on a clean install, and the same thing happened.