[REF] Useful init.d scripts - AT&T Samsung Galaxy S II SGH-I777

In the process of trying to improve my kernel, I've found that some of the potential improvements available for various kernels are sysfs tweaks that can be achieved in init.d - Some of these are things I was semi-aware of but didn't have the time to play around with until recently.
I'm going to start an archive of various scripts, categorized by "safe", "should be OK", and "dangerous but high payoff".
This post will link to individual posts in the thread, each with their own script.
For short scripts, I will simply put them here as code that can be copied and pasted, allowing for easier review and discussion. For longer scripts, I will attach them. I have started including downloadable versions of each script to avoid CR/LF issues. It is not possible to upload an attachment without an extension - before installing any attached scripts, you must remove the .txt extension
To install a script:
Put the contents into a file. I suggest an editor that can save in UNIX format (different linefeeds than DOS...)
Save the file - Save it without any extension. e.g. "governor_ondemand", not "governor_ondemand.sh"
Push the file as follows:
Code:
adb remount
adb push <file> /system/etc/init.d/<file>
adb chmod 755 /system/etc/init.d/<file>
The current script list:
Setting a CPU governor on boot
Remount all partitions with noatime
Disable per-file fsync()
Tweak I/O scheduler
Increase ext4 commit interval
Filesystem cache settings
Enabling AFTR (Exynos Only)
GPU Clock/Voltage Control (Exynos Only)

---

Setting a CPU governor on boot
This one's simple, and mainly for ROM developers. It sets the CPU governor as desired upon bootup. Most users will just use SetCPU instead.
Safety category: SAFE
Code:
#!/sbin/sh
log "Changing governor to ondemand"
echo "ondemand" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
Replace "ondemand" with the supported governor of your choice

---

Remount all partitions with noatime
One feature of ext4 is an "access time" attribute per file, aka atime. This attribute is not useful to very many people (mainly, I think, those doing security audits of sensitive systems). Updating the access time can slow filesystem access down.
The following script (Credit goes to gokhanmoral of SiyahKernel for this one) will remount all partitions with noatime set.
Safety category: SAFE
Code:
#!/sbin/sh
for k in $(/sbin/busybox mount | /sbin/busybox grep relatime | /sbin/busybox cut -d " " -f3)
do
sync
/sbin/busybox mount -o remount,noatime $k
done
Edit: garyd9 has pointed out that this may not actually be very beneficial - /data is already mounted noatime, doesn't matter on /system since that's read-only.

---

Enabling AFTR
AFTR is a deep-idle state supported by Exynos CPUs - it turns out that it is disabled by default. This script enables it. Credit to gokhanmoral of SiyahKernel for discovering this.
It is somewhat disconcerting that this feature is disabled by default and prints out a lot of debugging info by default. It may have issues. As a result,
Safety Category: SHOULD BE OK
Code:
#!/sbin/sh
log "Enabling AFTR"
echo "3" > /sys/module/cpuidle/parameters/enable_mask
Note that AFTR will cause your dmesg log to be spammed quite a bit by default. Daily Driver 12/8/2011 and newer have a fix for this, as do newer SiyahKernel releases.
This script is OBSOLETE on Daily Driver 1/29/2012 and later releases. AFTR is enabled by default on them.

---

Disable per-file fsync()
This capability is currently only supported in my kernel and (I think) knzo's void series for the I9100. It is based on patches from Andrea Righi (arighi). It disables per-file sync(), so that even if an application asks for data to be immediately flushed to disk, it won't be. This is most obvious with sqlite - this achieves the same goals as the modded sqlite you see in some tweaks, but in a runtime-controllable fashion.
However, this means that in a crash, you are much more likely to have unwritten database data. This can do things like cause a database to become corrupt, leading to force close loops. I had it happen all the time on xdandroid due to the way xdandroid handled filesystem writes on loop mounts.
Safety category: DANGEROUS (but potentially high payoff in database I/O workloads - this one is good for around 600 points of Quadrant epeen, for example)
Code:
#!/sbin/sh
log "Disabling per-file fsync()"
echo "1" > /sys/module/sync/parameters/fsync_disabled

---

Tweak I/O scheduler
This changes filesystem I/O scheduler parameters, and also changes the I/O scheduler if desired. This was included in many Infuse kernels, and is currently included in Daily Driver - However, as I believe that the user should have as much control as possible, this script will be removed from DD in favor of living "outside the kernel" where it should be within 1-2 releases.
While I discovered it in the Infuse community, its origin appears to be Bonsai.
Right now it defaults to BFQ. Change the "SCHEDULER=bfq" line to noop, deadline, or cfq as desired.
Safety category: SAFE (After all, it's been in DD almost since the beginning)
Code:
#!/sbin/sh
#
# Copyright (C) 2011 Free Software Foundation, Inc.
#
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU Library General Public License as published
# by the Free Software Foundation; either version 2, or (at your option)
# any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU Library General Public
# License along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
# USA.
#
# License GPLv2+: GNU GPL version 2 or later <http://gnu.org/licenses/gpl.html>
#
#################################
# BonsaiROM
#################################
MMC=`ls -d /sys/block/mmc*`
SCHEDULER=bfq
# Optimize non-rotating storage;
log "INIT.scheduler BEGIN:setting scheduler parameters for $SCHEDULER"
case "$SCHEDULER" in
noop)
for i in $MMC; do
echo "noop" > $i/queue/scheduler
done
;;
deadline)
for i in $MMC; do
echo "deadline" > $i/queue/scheduler
echo 80 > $i/queue/iosched/read_expire # untested
echo 1 > $i/queue/iosched/fifo_batch # untested
echo 1 > $i/queue/iosched/front_merges # untested
done
;;
cfq)
for i in $MMC; do
echo "cfq" > $i/queue/scheduler
echo "0" > $i/queue/rotational
echo "1" > $i/queue/iosched/back_seek_penalty
echo "1" > $i/queue/iosched/low_latency
echo "3" > $i/queue/iosched/slice_idle
echo "4096" > $i/queue/read_ahead_kb # default: 128; (recomended: 128)
echo 1000000000 > $i/queue/iosched/back_seek_max
echo "16" > $i/queue/iosched/quantum # default: 4 (recomended: 16)
echo "2048" > $i/queue/nr_requests # default:128 (recomended: 2048)
done
echo "0" > /proc/sys/kernel/sched_child_runs_first
;;
bfq)
for i in $MMC; do
echo "bfq" > $i/queue/scheduler
echo "0" > $i/queue/rotational
echo "1" > $i/queue/iosched/back_seek_penalty
echo "3" > $i/queue/iosched/slice_idle
echo "2048" > $i/queue/read_ahead_kb # default: 128
echo 1000000000 > $i/queue/iosched/back_seek_max
echo "16" > $i/queue/iosched/quantum # default: 4 (recomended: 16)
echo "2048" > $i/queue/nr_requests # default:128 (recomended: 2048)
done
;;
*)
log "INIT.scheduler ERROR:failed to set parameters for unknown scheduler: $SCHEDULER"
exit -1;
;;
esac
log "INIT.scheduler END:setting $SCHEDULER scheduler parameters"

---

Increase ext4 commit interval
This script increases the ext4 commit interval to 20 seconds (Default: 5) - by delaying journal writes longer, it can help performance, but lengthens the window during which a crash will cause unwritten data to be lost.
Credit goes to gokhanmoral of SiyahKernel for this one
Safety category: SHOULD BE OK
Code:
#!/sbin/sh
log "Increasing ext4 commit interval to 20 seconds
for k in $(/sbin/busybox mount | /sbin/busybox grep ext4 | /sbin/busybox cut -d " " -f3)
do
sync
/sbin/busybox mount -o remount,commit=20 $k
done

---

Filesystem cache settings
This one will be an external link. There are choices - battery-optimized or performance-optimized.
http://www.android-devs.com/?p=31
Safety category: SAFE

---

Excellent write up, thanks for sharing. Just a note, I think you typoed post #4 from post #2 (the code part).
Ken

---

Entropy512 said:
One feature of ext4 is an "access time" attribute per file, aka atime. This attribute is not useful to very many people (mainly, I think, those doing security audits of sensitive systems). Updating the access time can slow filesystem access down.
The following script (Credit goes to gokhanmoral of SiyahKernel for this one) will remount all partitions with noatime set.
Safety category: SAFE
Code:
#!/sbin/sh
for k in $(/sbin/busybox mount | /sbin/busybox grep relatime | /sbin/busybox cut -d " " -f3)
do
sync
/sbin/busybox mount -o remount,noatime $k
done
Click to expand...
Click to collapse
Thank you soo much for this script, Entropy!
ran it and noticed within moments ....
guys in the forum just a heads up ... in case you hadn't thought of it ... for ease of managing these scripts, you can download rom toolbox pro and enter a script (i think they're here for us to copy straight in to Terminal() + save new scripts etc etc. very convenient

---

Entropy512 said:
One feature of ext4 is an "access time" attribute per file, aka atime....
Click to expand...
Click to collapse
I'm surprised this is not off by default in Android. noatime is one of the first things recommended to turn off on file-operation-busy unix servers...

---

*redundancy edit*

---

Entropy512 said:
AFTR is a deep-idle state supported by our CPU....
Code:
#!/sbin/sh
log "Changing governor to ondemand"
echo "ondemand" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
...
Click to expand...
Click to collapse
the code above changes governor... is this a typo?
---------- Post added at 10:56 PM ---------- Previous post was at 10:47 PM ----------
Entropy512 said:
This script increases the ext4 commit interval to 20 seconds (Default: 5)
Click to expand...
Click to collapse
I wonder if anyone has confirmed real-life benefit of this? I mean it's a phone on (essentially) SSD drive with relatively little disk IO (especially on write) right?. It's not like we run Oracle on spindles where postponing commit longer may present kernel a chance to optimize/re-order IO operations with disks for better performance (but unsafer)....

---

vladm7 said:
I'm surprised this is not off by default in Android. noatime is one of the first things recommended to turn off on file-operation-busy unix servers...
Click to expand...
Click to collapse
Surprising

---

You can use terminal but.pretty sure They are for scripts in system/etc/init.d
You can use script manager if.your kernel does not support init.d
Amplified said:
Thank you soo much for this script, Entropy!
ran it and noticed within moments ....
guys in the forum just a heads up ... in case you hadn't thought of it ... for ease of managing these scripts, you can download rom toolbox pro and enter a script (i think they're here for us to copy straight in to Terminal() + save new scripts etc etc. very convenient
Click to expand...
Click to collapse
Sent from my SAMSUNG-SGH-I777 using xda premium

---

Entropy512 said:
This changes filesystem I/O scheduler parameters, and also changes the I/O scheduler if desired. This was included in many Infuse kernels, and is currently included in Daily Driver - However, as I believe that the user should have as much control as possible, this script will be removed from DD in favor of living "outside the kernel" where it should be within 1-2 releases.
While I discovered it in the Infuse community, its origin appears to be Bonsai.
Right now it defaults to BFQ. Change the "SCHEDULER=bfq" line to noop, deadline, or cfq as desired.
Safety category: SAFE (After all, it's been in DD almost since the beginning)
Code:
#!/sbin/sh
#
# Copyright (C) 2011 Free Software Foundation, Inc.
#
# This program is free software; you can redistribute it and/or modify it
# under the terms of the GNU Library General Public License as published
# by the Free Software Foundation; either version 2, or (at your option)
# any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU Library General Public
# License along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
# USA.
#
# License GPLv2+: GNU GPL version 2 or later <http://gnu.org/licenses/gpl.html>
#
#################################
# BonsaiROM
#################################
MMC=`ls -d /sys/block/mmc*`
SCHEDULER=bfq
# Optimize non-rotating storage;
log "INIT.scheduler BEGIN:setting scheduler parameters for $SCHEDULER"
case "$SCHEDULER" in
noop)
for i in $MMC; do
echo "noop" > $i/queue/scheduler
done
;;
deadline)
for i in $MMC; do
echo "deadline" > $i/queue/scheduler
echo 80 > $i/queue/iosched/read_expire # untested
echo 1 > $i/queue/iosched/fifo_batch # untested
echo 1 > $i/queue/iosched/front_merges # untested
done
;;
cfq)
for i in $MMC; do
echo "cfq" > $i/queue/scheduler
echo "0" > $i/queue/rotational
echo "1" > $i/queue/iosched/back_seek_penalty
echo "1" > $i/queue/iosched/low_latency
echo "3" > $i/queue/iosched/slice_idle
echo "4096" > $i/queue/read_ahead_kb # default: 128; (recomended: 128)
echo 1000000000 > $i/queue/iosched/back_seek_max
echo "16" > $i/queue/iosched/quantum # default: 4 (recomended: 16)
echo "2048" > $i/queue/nr_requests # default:128 (recomended: 2048)
done
echo "0" > /proc/sys/kernel/sched_child_runs_first
;;
bfq)
for i in $MMC; do
echo "bfq" > $i/queue/scheduler
echo "0" > $i/queue/rotational
echo "1" > $i/queue/iosched/back_seek_penalty
echo "3" > $i/queue/iosched/slice_idle
echo "2048" > $i/queue/read_ahead_kb # default: 128
echo 1000000000 > $i/queue/iosched/back_seek_max
echo "16" > $i/queue/iosched/quantum # default: 4 (recomended: 16)
echo "2048" > $i/queue/nr_requests # default:128 (recomended: 2048)
done
;;
*)
log "INIT.scheduler ERROR:failed to set parameters for unknown scheduler: $SCHEDULER"
exit -1;
;;
esac
log "INIT.scheduler END:setting $SCHEDULER scheduler parameters"
Click to expand...
Click to collapse
Installed this script verbatim, swapped bfq with the code noop (after a bit of reading up on these commands (within "Scheduler=bfq)) ~ and was unable to sucessfully run the script. Was wondering what else i should try to set scheduler to noop
*edit*
seperate mod, same useful init.d script thread oirigin
Also ~ after getting adb working i started attempting to tackle the sysctl.conf but after all a few hours of reading i have a couple of questions before i proceed to create this file / adb push it to my phone.
these are the only qs i have:
- Possible to create a file using windows? I assume possible, next q is how to create a *.conf file on a windows PC (at this hour i am not thinking outside of the box lol)
- once i create the sysctl.conf file, if i understand the process properly, i will simply add the codei chose into the file and then adb push the file to my phone which leads to couple sub questions (trying to learn - haters make like bee with no stinger): Should I simply cut / paste snipets of this code in whatever order ?
have it on my mind to complete this mod including new file - dev help is greatly appreciated.

---

Amplified said:
Surprising
Click to expand...
Click to collapse
well, yeah: i777 is my first Android.
Disabling atime is a widely used (safe) tuning in unix-servers world to increase performance of operations with the file system.
So, yes, I find it odd that Sammy didn't have that off by default in their ROMs.
Or am I missing the point of irony?

---

vladm7 said:
the code above changes governor... is this a typo?
Click to expand...
Click to collapse
I'll fix it tonight, that one's a copypasta error
vladm7 said:
I wonder if anyone has confirmed real-life benefit of this? I mean it's a phone on (essentially) SSD drive with relatively little disk IO (especially on write) right?. It's not like we run Oracle on spindles where postponing commit longer may present kernel a chance to optimize/re-order IO operations with disks for better performance (but unsafer)....
Click to expand...
Click to collapse
I haven't poked at that one much. It depends partly on if the phone will enter deep sleep with uncommitted data. It may only be good for benchmark epeen
As to scripts in /system/etc/init.d not running:
They must be executable (chmod 755 is recommended)
They must not have a .sh extension - for some reason run-parts won't run them if they do (annoying...)
If created in Windows there might be line-ending issues (CR vs CR/LF)

---

a couple of small scripts to remount in RW/RO
I know, this is not directly related to kernel params tuning, but I think not worth a new thread.
Since I often mess in terminal, and need to remount /system in RW mode, I wrote those two small scripts to save some extra typing.
Remounts given file system in RW, /system by default. Save as /system/xbin/rw and then chmod +rx /system/xbin/rw:
Code:
#!/sbin/sh
disk=/system
if [ -n "$1" ]; then
if [ $(busybox grep -c $1 /proc/mounts) -gt 0 ]; then
disk="$1"
else
echo "Can't find '$1' mount"
exit
fi
fi
echo "Remounting '$disk' in Read-Write access mode"
busybox mount -o rw,remount $disk
Save as /system/xbin/ro. Remounts given file system in RO, /system by default:
Code:
#!/sbin/sh
disk=/system
if [ -n "$1" ]; then
if [ $(busybox grep -c $1 /proc/mounts) -gt 0 ]; then
disk="$1"
else
echo "Can't find '$1' mount"
exit
fi
fi
echo "Remounting '$disk' in Read-Only access mode"
busybox mount -o ro,remount $disk

---

Entropy,
Can you attach these scripts please.