Const Coccinelle

TheCosmicStaff Kernel For M21

CONTENTS

Features

Changelog

R6:

R5-noboost:

  • Merge boost and noboost into one
  • Bring in some fixes to samsung drivers
  • Add support for Dynamic SchedTune Boost
  • Add support for dynamic prefer_idle
  • Stub out debug prints by default (Graphics perf improvement)
  • Reserve caches for small, high-frequency memory allocations (Reduce latency)
  • Use power efficient workingqueues for drm (Battery life improvement)
  • Use power efficient workingqueues for block (Battery life improvement)
  • Minimize number of tasks to load balance
  • Add Srandom (5 times faster than urandom)
  • Resolve sched_feat() at compile time to improve code optimization (Perf improvement)
  • Upstream to Linux 4.14.124
  • only force UX tasks to big cores (Power saving)
  • Disable UFS debugging (IO perf improvement)
  • Use SCHED_RR in place of SCHED_FIFO for all users (UX improvement)
  • Change default SCHED_RR timeslice from 100 ms to 1 jiffy (Lowered latency)
  • Make zsmalloc copy rather than mapping pages (Memory perf improvement)
  • Disable Selinux audit
  • Don’t dynamically allocate single-use structs (Reduce overhead)
  • Avoid dynamic memory allocation for small write buffers (Reduce overhead)
  • Avoid dynamically allocating memory in getxattr (Perf improvement)
  • Schedule workers on CPU0 or 0-5 by default (Power saving)
  • Fix memory leak for uncached firmware requests
  • Speed up cache entry creation (Reduce overhead)
  • Avoid dynamic allocations in sel_write_access (Perf improvement)
  • Avoid dynamic memory allocation for INITCONTEXTLEN buffers (Reduce overhead)
  • Use kmem_cache pool for struct dma_buf_attachment, dmabuf allocations, struct sync_file, sdcardfs_file_info, cgrp_cset_link, kernfs_open_node/file, async_entry (Reduce overhead)
  • Always compile core debugfs driver for Android kernels
  • Avoid allocating small buffers for map keys and values (Reduce overhead)
  • Start killing wakelocks after one minute of idle (Power saving)
  • Expedite garbage collection if idle (Power saving)
  • Add a timeout to wakelocks globally
  • Reduce QoS boosting from Samsung hacks (Power saving)
  • Disable full refcount validation (Perf improvement)
  • Micro-optimize PID map reads for arm64 while retaining output format (Help with games like genshin impact)
  • Stop kswapd early when nothing’s waiting for it to free pages
  • Don’t stop kswapd on a per-node basis when there are no waiters
  • Move frequently used functions to headers and declare them inline (Reduce overhead)
  • Don’t hog RCU read lock while optimistically spinning
  • Use interruptible wait for genirq, media v4l (Reduce latency)
  • Increase watermark scale factor
  • Update LZ4 components
  • Reduce latency while processing atomic ioctls (Reduce display rendering latency)
  • Drop top-app level to 2

R4-noboost:

  • Introduce SchedTune Assist (500% improvement in User experience at minimal cost of battery life)
  • Increase the time a task is considered cache-hot (Performance gain)
  • Queue requests on their origin CPU (20%-40% gain in some scenarios)
  • Do not give sleepers 50% more runtime
  • Do not wake idle CPUs to queue same-origin requests (noticeable battery life improvement)
  • Do not use IPIs for remote wakeups if idle
  • Allow aggressive remote task interruptions
  • Do not collect I/O statistics
  • Don’t allow userspace to impose restrictions on CPU idle levels (No more random performance drops)
  • Implement optimised checksum routine (Peformance gain)
  • Network Stack optimizations
  • Don’t hog RCU read lock while optimistically spinning
  • Speed up ioctl by omitting debug names (overall graphics performance boost)
  • optimized memcpy (Performance gain)
  • optimized memmove (Performance gain)
  • optimized memset (Performance gain)
  • Enable BPF JIT (Faster settings activities launch)
  • Update default wakelock block list (Better battery life)
  • Enable wireguard (add wireguard support for those who might need it)

R4-boost:

  • Don’t allow userspace to impose restrictions on CPU idle levels (No more random performance drops)
  • Implement optimised checksum routine (Peformance gain)
  • Network Stack optimizations
  • Don’t hog RCU read lock while optimistically spinning
  • Speed up ioctl by omitting debug names (overall graphics performance boost)
  • optimized memcpy (Performance gain)
  • optimized memmove (Performance gain)
  • optimized memset (Performance gain)
  • Enable BPF JIT (Faster settings activities launch)
  • Update default wakelock block list (Better battery life)
  • update boost freq (less heating)
  • Change default SCHED_RR timeslice from 100 ms to 1 jiffy
  • Use SCHED_RR in place of SCHED_FIFO for all users (lowered latency)
  • Micro-optimize PID map reads for arm64 while retaining output format (Should help with certain big games)
  • Free dead tasks asynchronously in finish_task_switch() (Performance gain)
  • Avoid allocating small buffers for map keys and values
  • Avoid dynamic memory allocation for small write buffers (avoid overhead)
  • Reduce latency while processing atomic ioctls (lowered latency)
  • Enable wireguard (add wireguard support for those who might need it)

R3:

  • drop Schedtune assist
  • drop dsboost
  • revert undervolting
  • set timer 300hz
  • Boost DDR bus for a short amount of time when zygote forks (faster app launches)
  • Use -O3 optimization (slight performance improvement)
  • branch optimization in free slowpath
  • Align file struct to 8 bytes
  • disable crc checks (30% IO speed improvement)
  • optimize modulo operation
  • set initial TCP window size to 64K (speed improvement)
  • use power efficient workingqueues
  • add NEON accelerated XOR implementation
  • set dirty ratio to 40%
  • Increase vmstat interval
  • swap pages one at a time (help with extreme memory pressure scenarios)

R2:

  • Fix error “There’s an internal problem with your system”
  • Enable Write Back Caching (improve flash storage lifespan)
  • Don’t fail on wakeup (do not let userspace abuse alarmtimers)
  • Reduce the opportunity for sleepers to preempt (reduce jitter)
  • Enable I/O throttling (reduce latency)
  • Introduce SchedTune Assist (faster app launches and power efficiency)
  • Introduce DSBoost driver(smoother experience and lowered jank)
  • Introduce devfreq boost driver (butter smooth UI experience, no frame-drops)
  • Optimized Console FrameBuffer for upto 70% increase in Performance
  • Boost DDR bus upon running an atomic ioctl (prevent any and all framedrops)
  • increase limit on sched-tune boost groups
  • Undervolt the whole soc (lowered power usage)
  • Process new forks before processing their parent (improve app launch speed)
  • apply boost before frames are updated
  • enable westwood (lowered network latency and better speeds)
  • Prefer to reclaim inode and dentry cache over pagecache
  • Increase ratelimit pages value (reduce overhead)

R1:

  • Tweak cpuidle governor to enter deep C state faster(better battery life)
  • set frequency of kernel interrupt to 100hz(noticeable battery life improvement)
  • Add boeffla Wakelock blocker
  • Disable debugging in binder(lowered latency)
  • Improve 3x faster integer sqrt
  • disable IO_STAT completely(IO speed improvement)
  • ARM64 targeted optimizations
  • Make workqueue high priority(Decrease delays under pressure)
  • cacheline alignment(Performance benefit)
  • Don’t let google crap run in the background(batter battery life)
  • Make zsmalloc copy rather than mapping pages(performance improvement)
  • Disable audit
  • Disable auditing for selinux

Flashing instructions

Downlods

Sources

Note

I built it for my personal use only, my goals were to achieve better efficiency and not performance

Updated on