tuned/functions

#
# This is library of helper functions that can be used in scripts in tuned profiles.
#
# API provided by this library is under heavy development and could be changed anytime
#

#
# Config
#
STORAGE=/run/tuned
STORAGE_PERSISTENT=/var/lib/tuned
STORAGE_SUFFIX=".save"

#
# Helpers
#

# Save value
# $0 STORAGE_NAME VALUE
save_value() {
	[ "$#" -ne 2 ] && return
	[ "$2" -a -e "${STORAGE}" ] && echo "$2" > "${STORAGE}/${1}${STORAGE_SUFFIX}"
}

# Parse sysfs value, i.e. for "val1 [val2] val3" return "val2"
# $0 SYSFS_NAME
parse_sys() {
	local V1 V2
	[ -r "$1" ] || return
	V1=`cat "$1"`
	V2="${V1##*[}"
	V2="${V2%%]*}"
	echo "${V2:-$V1}"
}

# Save sysfs value
# $0 STORAGE_NAME SYSFS_NAME
save_sys() {
	[ "$#" -ne 2 ] && return
	[ -r "$2" -a ! -e "${STORAGE}/${1}${STORAGE_SUFFIX}" ] && parse_sys "$2" > "${STORAGE}/${1}${STORAGE_SUFFIX}"
}

# Set sysfs value
# $0 SYSFS_NAME VALUE
set_sys() {
	[ "$#" -ne 2 ] && return
	[ -w "$1" ] && echo "$2" > "$1"
}

# Save and set sysfs value
# $0 STORAGE_NAME SYSFS_NAME VALUE
save_set_sys() {
	[ "$#" -ne 3 ] && return
	save_sys "$1" "$2"
	set_sys "$2" "$3"
}

# Get stored sysfs value from storage
# $0 STORAGE_NAME
get_stored_sys() {
	[ "$#" -ne 1 ] && return
	[ -r "${STORAGE}/${1}${STORAGE_SUFFIX}" ] && cat "${STORAGE}/${1}${STORAGE_SUFFIX}"
}

# Restore value from storage
# $0 STORAGE_NAME
restore_value() {
	[ "$#" -ne 1 ] && return
	_rs_value="`get_stored_sys \"$1\"`"
	unlink "${STORAGE}/${1}${STORAGE_SUFFIX}" >/dev/null 2>&1
	[ "$_rs_value" ] && echo "$_rs_value"
}

# Restore sysfs value from storage, if nothing is stored, use VALUE
# $0 STORAGE_NAME SYSFS_NAME [VALUE]
restore_sys() {
	[ "$#" -lt 2 -o "$#" -gt 3 ] && return
	_rs_value="`get_stored_sys \"$1\"`"
	unlink "${STORAGE}/${1}${STORAGE_SUFFIX}" >/dev/null 2>&1
	[ "$_rs_value" ] || _rs_value="$3"
	[ "$_rs_value" ] && set_sys "$2" "$_rs_value"
}


#
# DISK tuning
#

DISKS_DEV="$(command ls -d1 /dev/[shv]d*[a-z] 2>/dev/null)"
DISKS_SYS="$(command ls -d1 /sys/block/{sd,cciss,dm-,vd,dasd,xvd}* 2>/dev/null)"

_check_elevator_override()
{
	/bin/fgrep -q 'elevator=' /proc/cmdline
}

# $0 OPERATOR DEVICES ELEVATOR
_set_elevator_helper() {
	_check_elevator_override && return
	SYS_BLOCK_SDX=""
	[ "$2" ] && SYS_BLOCK_SDX=$(eval LANG=C /bin/ls -1 "${2}" 2>/dev/null)

	# if there is no kernel command line elevator settings, apply the elevator
	if [ "$1" -a "$SYS_BLOCK_SDX" ]; then
		for i in $SYS_BLOCK_SDX; do
			se_dev="`echo \"$i\" | sed 's|/sys/block/\([^/]\+\)/queue/scheduler|\1|'`"
			$1 "elevator_${se_dev}" "$i" "$3"
		done
	fi
}

# $0 DEVICES ELEVATOR
set_elevator() {
	_set_elevator_helper save_set_sys "$1" "$2"
}

# $0 DEVICES [ELEVATOR]
restore_elevator() {
	re_elevator="$2"
	[ "$re_elevator" ] || re_elevator=cfq
	_set_elevator_helper restore_sys "$1" "$re_elevator"
}

# SATA Aggressive Link Power Management
# usage: set_disk_alpm policy
set_disk_alpm() {
	policy=$1

        for host in /sys/class/scsi_host/*; do
                if [ -f $host/ahci_port_cmd ]; then
                        port_cmd=`cat $host/ahci_port_cmd`;
                        if [ $((0x$port_cmd & 0x240000)) = 0 -a -f $host/link_power_management_policy ]; then
                                echo $policy >$host/link_power_management_policy;
                        else
                                echo "max_performance" >$host/link_power_management_policy;
                        fi
                fi
        done
}

# usage: set_disk_apm level
set_disk_apm() {
	level=$1
	for disk in $DISKS_DEV; do
		hdparm -B $level $disk &>/dev/null
	done
}

# usage: set_disk_spindown level
set_disk_spindown() {
	level=$1
	for disk in $DISKS_DEV; do
		hdparm -S $level $disk &>/dev/null
	done
}

# usage: multiply_disk_readahead by
multiply_disk_readahead() {
	by=$1

	# float multiplication not supported in bash
	# bc might not be installed, python is available for sure

	for disk in $DISKS_SYS; do
		control="${disk}/queue/read_ahead_kb"
		old=$(cat $control)
		new=$(echo "print int($old*$by)" | python)

		(echo $new > $control) &>/dev/null
	done
}

# usage: remount_disk options partition1 partition2 ...
remount_partitions() {
	options=$1
	shift

	for partition in $@; do
		mount -o remount,$options $partition >/dev/null 2>&1
	done
}

remount_all_no_rootboot_partitions() {
	[ "$1" ] || return
	# Find non-root and non-boot partitions, disable barriers on them
	rootvol=$(df -h / | grep "^/dev" | awk '{print $1}')
	bootvol=$(df -h /boot | grep "^/dev" | awk '{print $1}')
	volumes=$(df -hl --exclude=tmpfs | grep "^/dev" | awk '{print $1}')
	nobarriervols=$(echo "$volumes" | grep -v $rootvol | grep -v $bootvol)
	remount_partitions "$1" $nobarriervols
}


DISK_QUANTUM_SAVE="${STORAGE}/disk_quantum${STORAGE_SUFFIX}"

set_disk_scheduler_quantum() {
	value=$1
	rm -f "$DISK_QUANTUM_SAVE"
	for disk in $DISKS_SYS; do
		control="${disk}/queue/iosched/quantum"
		echo "echo $(cat $control) > $control" >> "$DISK_QUANTUM_SAVE" 2>/dev/null
		(echo $value > $control) &2>/dev/null
	done
}

restore_disk_scheduler_quantum() {
	if [ -r "$DISK_QUANTUM_SAVE" ]; then
		/bin/bash "$DISK_QUANTUM_SAVE" &>/dev/null
		rm -f "$DISK_QUANTUM_SAVE"
	fi
}

#
# CPU tuning
#

CPUSPEED_ORIG_GOV="${STORAGE}/cpuspeed-governor-%s${STORAGE_SUFFIX}"
CPUS="$(ls -d1 /sys/devices/system/cpu/cpu* | sed 's;^.*/;;' |  grep "cpu[0-9]\+")"

# set CPU governor setting and store the old settings
# usage: set_cpu_governor governor
set_cpu_governor() {
	governor=$1

	# direct change using sysfs
	if [ -e /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor ]; then

		for cpu in $CPUS; do
			gov_file=/sys/devices/system/cpu/$cpu/cpufreq/scaling_governor
			save_file=$(printf $CPUSPEED_ORIG_GOV $cpu)
			rm -f $save_file
			if [ -e $gov_file ]; then
				cat $gov_file > $save_file
				echo $governor > $gov_file
			fi
		done
	fi
}

# re-enable previous CPU governor settings
# usage: restore_cpu_governor
restore_cpu_governor() {
	if [ -e /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor ]; then
		for cpu in $CPUS; do
			cpufreq_dir=/sys/devices/system/cpu/$cpu/cpufreq
			save_file=$(printf $CPUSPEED_ORIG_GOV $cpu)

			if [ -e $cpufreq_dir/scaling_governor ]; then
				if [ -e $save_file ]; then
					cat $save_file > $cpufreq_dir/scaling_governor
					rm -f $save_file
				else
					echo userspace > $cpufreq_dir/scaling_governor
					cat $cpufreq_dir/cpuinfo_max_freq > $cpufreq_dir/scaling_setspeed
				fi
			fi
		done
	fi
}

_cpu_multicore_powersave() {
	value=$1
	[ -e /sys/devices/system/cpu/sched_mc_power_savings ] && echo $value > /sys/devices/system/cpu/sched_mc_power_savings
}

# enable multi core power savings for low wakeup systems
enable_cpu_multicore_powersave() {
	_cpu_multicore_powersave 1
}

disable_cpu_multicore_powersave() {
	_cpu_multicore_powersave 0
}

#
# MEMORY tuning
#

THP_ENABLE="/sys/kernel/mm/transparent_hugepage/enabled"
THP_SAVE="${STORAGE}/thp${STORAGE_SUFFIX}"

[ -e "$THP_ENABLE" ] || THP_ENABLE="/sys/kernel/mm/redhat_transparent_hugepage/enabled"

enable_transparent_hugepages() {
	if [ -e $THP_ENABLE ]; then
		cut -f2 -d'[' $THP_ENABLE  | cut -f1 -d']' > $THP_SAVE
		(echo always > $THP_ENABLE) &> /dev/null
	fi
}

restore_transparent_hugepages() {
	if [ -e $THP_SAVE ]; then
		(echo $(cat $THP_SAVE) > $THP_ENABLE) &> /dev/null
		rm -f $THP_SAVE
	fi
}

#
# WIFI tuning
#

# usage: _wifi_set_power_level level
_wifi_set_power_level() {
	# 0    auto, PM enabled
	# 1-5  least savings and lowest latency - most savings and highest latency
	# 6    disable power savings
	level=$1

	# do not report errors on systems with no wireless
	[ -e /proc/net/wireless ] || return 0

	# apply the settings using iwpriv
	ifaces=$(cat /proc/net/wireless | grep -v '|' | sed 's@^ *\([^:]*\):.*@\1@')
	for iface in $ifaces; do
		iwpriv $iface set_power $level
	done

	# some adapters may rely on sysfs
	for i in /sys/bus/pci/devices/*/power_level; do
		(echo $level > $i) &> /dev/null
	done
}

enable_wifi_powersave() {
	_wifi_set_power_level 5
}

disable_wifi_powersave() {
	_wifi_set_power_level 0
}

#
# BLUETOOTH tuning
#

disable_bluetooth() {
	hciconfig hci0 down >/dev/null 2>&1
	lsmod | grep -q btusb && rmmod btusb
}

enable_bluetooth() {
	modprobe btusb
	hciconfig hci0 up >/dev/null 2>&1
}

#
# USB tuning
#

_usb_autosuspend() {
	value=$1
	for i in /sys/bus/usb/devices/*/power/autosuspend; do echo $value > $i; done &> /dev/null
}

enable_usb_autosuspend() {
	_usb_autosuspend 1
}

disable_usb_autosuspend() {
	_usb_autosuspend 0
}

#
# SOUND CARDS tuning
#

enable_snd_ac97_powersave() {
	save_set_sys ac97 /sys/module/snd_ac97_codec/parameters/power_save Y
}

disable_snd_ac97_powersave() {
	save_set_sys ac97 /sys/module/snd_ac97_codec/parameters/power_save N
}

restore_snd_ac97_powersave() {
	restore_sys ac97 /sys/module/snd_ac97_codec/parameters/power_save $1
}

set_hda_intel_powersave() {
	save_set_sys hda_intel /sys/module/snd_hda_intel/parameters/power_save $1
}

restore_hda_intel_powersave() {
	restore_sys hda_intel /sys/module/snd_hda_intel/parameters/power_save $1
}

#
# VIDEO CARDS tuning
#

# Power savings settings for Radeon
# usage: set_radeon_powersave dynpm | default | low | mid | high
set_radeon_powersave () {
	[ "$1" ] || return
	[ -e /sys/class/drm/card0/device/power_method ] || return
	if [ "$1" = default -o "$1" = auto -o "$1" = low -o "$1" = med -o "$1" = high ]; then
		[ -w /sys/class/drm/card0/device/power_profile ] || return
		save_sys radeon_profile /sys/class/drm/card0/device/power_profile
		save_set_sys radeon_method /sys/class/drm/card0/device/power_method profile
		set_sys /sys/class/drm/card0/device/power_profile "$1"
	elif [ "$1" = dynpm ]; then
		save_sys radeon_profile /sys/class/drm/card0/device/power_profile
		save_set_sys radeon_method /sys/class/drm/card0/device/power_method dynpm
	fi
}

restore_radeon_powersave () {
  restore_sys radeon_method /sys/class/drm/card0/device/power_method profile
  _rrp_method="`get_stored_sys radeon_method`"
  [ -z "$_rrp_method" -o _rrp_method="profile" ] && restore_sys radeon_profile /sys/class/drm/card0/device/power_profile default
}

#
# SOFTWARE tuning
#

RSYSLOG_CFG="/etc/rsyslog.conf"
RSYSLOG_SAVE="${STORAGE}/cpuspeed${STORAGE_SUFFIX}"

disable_logs_syncing() {
	cp -p $RSYSLOG_CFG $RSYSLOG_SAVE
	sed -i 's/ \/var\/log/-\/var\/log/' $RSYSLOG_CFG
}

restore_logs_syncing() {
	mv -Z $RSYSLOG_SAVE $RSYSLOG_CFG || mv $RSYSLOG_SAVE $RSYSLOG_CFG
}

#
# HARDWARE SPECIFIC tuning
#

# Asus EEE with Intel Atom
_eee_fsb_control() {
	value=$1
	if [ -e /sys/devices/platform/eeepc/she ]; then
		echo $value > /sys/devices/platform/eeepc/she
	elif [ -e /sys/devices/platform/eeepc/cpufv ]; then
		echo $value > /sys/devices/platform/eeepc/cpufv
	elif [ -e /sys/devices/platform/eeepc-wmi/cpufv ]; then
		echo $value > /sys/devices/platform/eeepc-wmi/cpufv
	fi
}

eee_set_reduced_fsb() {
	_eee_fsb_control 2
}

eee_set_normal_fsb() {
	_eee_fsb_control 1
}

#
# modprobe configuration handling
#

kvm_modprobe_file=/etc/modprobe.d/kvm.rt.tuned.conf

teardown_kvm_mod_low_latency()
{
	rm -f $kvm_modprobe_file
}

setup_kvm_mod_low_latency()
{
	local HAS_KPS=""
	local HAS_NX_HP=""
	local HAS_PLE_GAP=""
	local WANTS_KPS=""
	local WANTS_NX_HP=""
	local WANTS_PLE_GAP=""

        if ! modinfo kvm >/dev/null; then
            teardown_kvm_mod_low_latency
            return 0
        fi

	modinfo -p kvm | grep -q kvmclock_periodic_sync && HAS_KPS=1
	modinfo -p kvm | grep -q nx_huge_pages && HAS_NX_HP=1
	modinfo -p kvm_intel 2>/dev/null | grep -q ple_gap && HAS_PLE_GAP=1
	grep -qs kvmclock_periodic_sync "$kvm_modprobe_file" && WANTS_KPS=1
	grep -qs nx_huge_pages "$kvm_modprobe_file" && WANTS_NX_HP=1
	grep -qs ple_gap "$kvm_modprobe_file" && WANTS_PLE_GAP=1

	if [ "$HAS_KPS" != "$WANTS_KPS" -o "$HAS_PLE_GAP" != "$WANTS_PLE_GAP" -o \
	     "$HAS_NX_HP" != "$WANTS_NX_HP" ]; then
		teardown_kvm_mod_low_latency
		[ "$HAS_KPS" ] && echo "options kvm kvmclock_periodic_sync=0" > $kvm_modprobe_file
		[ "$HAS_NX_HP" ] && echo "options kvm nx_huge_pages=0" >> $kvm_modprobe_file
		[ "$HAS_PLE_GAP" ] && echo "options kvm_intel ple_gap=0" >> $kvm_modprobe_file
	fi
	return 0
}

#
# KSM
#

KSM_SERVICES="ksm ksmtuned"
KSM_RUN_PATH=/sys/kernel/mm/ksm/run
KSM_MASK_FILE="${STORAGE_PERSISTENT}/ksm-masked"

disable_ksm()
{
	if [ ! -f $KSM_MASK_FILE ]; then
		# Always create $KSM_MASK_FILE, since we don't want to
		# run any systemctl commands during boot
		if ! touch $KSM_MASK_FILE; then
			die "failed to create $KSM_MASK_FILE"
		fi
		# Do not run any systemctl commands if $KSM_SERVICES units do not exist
		systemctl cat -- $KSM_SERVICES &> /dev/null || return 0
		systemctl --now --quiet mask $KSM_SERVICES
		# Unmerge all shared pages
		test -f $KSM_RUN_PATH && echo 2 > $KSM_RUN_PATH
	fi
}

# Should only be called when full_rollback == true
enable_ksm()
{
	if [ -f $KSM_MASK_FILE ]; then
		# Do not run any systemctl commands if $KSM_SERVICES units do not exist
		systemctl cat -- $KSM_SERVICES &> /dev/null || return 0
		if systemctl --quiet unmask $KSM_SERVICES; then
			rm -f $KSM_MASK_FILE
		fi
	fi
}

die() {
	echo "$@" >&2
	exit 1
}

#
# ACTION PROCESSING
#

error_not_implemented() {
	echo "tuned: script function '$1' is not implemented." >&2
}

# implicit actions, will be used if not provided by profile script:
#
# * start    must be implemented
# * stop     must be implemented

start() {
	error_not_implemented start
	return 16
}

stop() {
	error_not_implemented stop
	return 16
}

#
# main processing
#

process() {
	ARG="$1"
	shift
	case "$ARG" in
	start)
		start "$@"
		RETVAL=$?
		;;
	stop)
		stop "$@"
		RETVAL=$?
		;;
	verify)
		if declare -f verify &> /dev/null;
		then
			verify "$@"
		else
			:
		fi
		RETVAL=$?
		;;
	*)
		echo $"Usage: $0 {start|stop|verify}"
		RETVAL=2
		;;
	esac

	exit $RETVAL
}