mcreboot/mcstop+release: support for regular user execution
Change-Id: I9088f9c49bea13826bbab6348aa5560e6d91071b
This commit is contained in:
Balazs Gerofi
committed by
Dominique Martinet
Dominique Martinet
parent
13758417c5
commit
2dc51530f3
@@ -1,9 +1,9 @@
|
||||
#!/bin/bash
|
||||
# mcreboot-smp-x86.sh.in COPYRIGHT FUJITSU LIMITED 2018
|
||||
|
||||
# IHK SMP-x86 example boot script.
|
||||
# IHK/McKernel SMP boot script.
|
||||
# author: Balazs Gerofi <bgerofi@riken.jp>
|
||||
# Copyright (C) 2014 RIKEN AICS
|
||||
# modified: Fujitsu Ltd.
|
||||
# Copyright (C) 2018
|
||||
#
|
||||
# This is an example script for loading IHK, configuring a partition and
|
||||
# booting McKernel on it. Unless specific CPUs and memory are requested,
|
||||
@@ -12,6 +12,11 @@
|
||||
# Otherwise, it destroys the current McKernel instance and reboots it using
|
||||
# the same set of resources as it used previously.
|
||||
# Note that the script does not output anything unless an error occurs.
|
||||
#
|
||||
# This script is meant to be run under sudo. Running it with regular user
|
||||
# privileges requires /etc/sudoers to contain the following persmissions:
|
||||
#
|
||||
# hpcuser ALL=(root) NOPASSWD: /sbin/insmod, /sbin/rmmod, /sbin/setenforce, /bin/systemctl, /bin/cat, /bin/cp /tmp/mcreboot/smp_affinity /proc/irq/*/smp_affinity, /bin/chown * /dev/mcd*, /bin/chown * /dev/mcos*, /*/*/*/*/ihkconfig, /*/*/*/*/ihkosctl, /usr/bin/tee, /bin/sync
|
||||
|
||||
ret=1
|
||||
prefix="@prefix@"
|
||||
@@ -41,7 +46,11 @@ redirect_kmsg=0
|
||||
mon_interval="-1"
|
||||
DUMP_LEVEL=24
|
||||
facility="LOG_LOCAL6"
|
||||
chown_option=`logname 2> /dev/null`
|
||||
chown_option=`id -un 2> /dev/null`
|
||||
# For the case when the entire script is run with sudo fall back to login name
|
||||
if [ "${chown_option}" == "root" ]; then
|
||||
chown_option=`logname 2> /dev/null`
|
||||
fi
|
||||
|
||||
if [ "`systemctl status irqbalance_mck.service 2> /dev/null |grep -E 'Active: active'`" != "" -o "`systemctl status irqbalance.service 2> /dev/null |grep -E 'Active: active'`" != "" ]; then
|
||||
irqbalance_used="yes"
|
||||
@@ -112,7 +121,7 @@ fi
|
||||
|
||||
disable_irqbalance_mck() {
|
||||
if [ -f /etc/systemd/system/irqbalance_mck.service ]; then
|
||||
systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null
|
||||
sudo systemctl disable irqbalance_mck.service >/dev/null 2>/dev/null
|
||||
|
||||
# Invalid .service file persists so remove it
|
||||
rm -f /etc/systemd/system/irqbalance_mck.service
|
||||
@@ -129,10 +138,11 @@ error_exit() {
|
||||
irqbalance_mck_started)
|
||||
if [ "${irqbalance_used}" == "yes" ]; then
|
||||
if [ "`systemctl status irqbalance_mck.service 2> /dev/null |grep -E 'Active: active'`" != "" ]; then
|
||||
if ! systemctl stop irqbalance_mck.service 2>/dev/null; then
|
||||
if ! sudo systemctl stop irqbalance_mck.service 2>/dev/null; then
|
||||
echo "warning: failed to stop irqbalance_mck" >&2
|
||||
fi
|
||||
disable_irqbalance_mck
|
||||
rm /tmp/irqbalance_mck
|
||||
fi
|
||||
fi
|
||||
;&
|
||||
@@ -141,36 +151,36 @@ error_exit() {
|
||||
if ls /dev/mcos* 1>/dev/null 2>&1; then
|
||||
for i in /dev/mcos*; do
|
||||
ind=`echo $i|cut -c10-`;
|
||||
if ! ${SBINDIR}/ihkconfig 0 destroy $ind; then
|
||||
if ! sudo ${SBINDIR}/ihkconfig 0 destroy $ind; then
|
||||
echo "warning: failed to destroy LWK instance $ind" >&2
|
||||
fi
|
||||
done
|
||||
fi
|
||||
;&
|
||||
mcctrl_loaded)
|
||||
rmmod mcctrl 2>/dev/null || echo "warning: failed to remove mcctrl" >&2
|
||||
sudo rmmod mcctrl 2>/dev/null || echo "warning: failed to remove mcctrl" >&2
|
||||
;&
|
||||
cpus_reserved)
|
||||
cpus=`${SBINDIR}/ihkconfig 0 query cpu`
|
||||
cpus=`sudo ${SBINDIR}/ihkconfig 0 query cpu`
|
||||
if [ "${cpus}" != "" ]; then
|
||||
if ! ${SBINDIR}/ihkconfig 0 release cpu $cpus > /dev/null; then
|
||||
if ! sudo ${SBINDIR}/ihkconfig 0 release cpu $cpus > /dev/null; then
|
||||
echo "warning: failed to release CPUs" >&2
|
||||
fi
|
||||
fi
|
||||
;&
|
||||
mem_reserved)
|
||||
mem=`${SBINDIR}/ihkconfig 0 query mem`
|
||||
mem=`sudo ${SBINDIR}/ihkconfig 0 query mem`
|
||||
if [ "${mem}" != "" ]; then
|
||||
if ! ${SBINDIR}/ihkconfig 0 release mem $mem > /dev/null; then
|
||||
if ! sudo ${SBINDIR}/ihkconfig 0 release mem $mem > /dev/null; then
|
||||
echo "warning: failed to release memory" >&2
|
||||
fi
|
||||
fi
|
||||
;&
|
||||
ihk_smp_loaded)
|
||||
rmmod ihk_smp_@ARCH@ 2>/dev/null || echo "warning: failed to remove ihk_smp_@ARCH@" >&2
|
||||
sudo rmmod ihk_smp_@ARCH@ 2>/dev/null || echo "warning: failed to remove ihk_smp_@ARCH@" >&2
|
||||
;&
|
||||
ihk_loaded)
|
||||
rmmod ihk 2>/dev/null || echo "warning: failed to remove ihk" >&2
|
||||
sudo rmmod ihk 2>/dev/null || echo "warning: failed to remove ihk" >&2
|
||||
;&
|
||||
smp_affinity_modified)
|
||||
umask $umask_old
|
||||
@@ -183,7 +193,7 @@ error_exit() {
|
||||
;&
|
||||
irqbalance_stopped)
|
||||
if [ "${irqbalance_used}" == "yes" ]; then
|
||||
if ! systemctl start irqbalance.service; then
|
||||
if ! sudo systemctl start irqbalance.service; then
|
||||
echo "warning: failed to start irqbalance" >&2;
|
||||
fi
|
||||
fi
|
||||
@@ -225,13 +235,13 @@ fi
|
||||
|
||||
# Stop irqbalance
|
||||
if [ "${irqbalance_used}" == "yes" ]; then
|
||||
systemctl stop irqbalance_mck.service 2>/dev/null
|
||||
if ! systemctl stop irqbalance.service 2>/dev/null ; then
|
||||
sudo systemctl stop irqbalance_mck.service 2>/dev/null
|
||||
if ! sudo systemctl stop irqbalance.service 2>/dev/null ; then
|
||||
echo "error: stopping irqbalance" >&2
|
||||
exit 1
|
||||
fi;
|
||||
|
||||
if ! perl -e 'use File::Copy qw(copy); $tmpdir="/tmp/mcreboot"; @files = grep { -f } glob "/proc/irq/*/smp_affinity"; foreach $file (@files) { $rel = substr($file, 1); $dir = substr($rel, 0, length($rel) - length("/smp_affinity")); if (system("mkdir -p $tmpdir/$dir")) { exit 1; } if (0) { print "cp $file $tmpdir/$rel\n"; } if (!copy($file,"$tmpdir/$rel")) { exit 1; } }'; then
|
||||
if ! perl -e 'use File::Copy qw(copy); $tmpdir="/tmp/mcreboot"; @files = grep { -f } glob "/proc/irq/*/smp_affinity"; foreach $file (@files) { $rel = substr($file, 1); $dir = substr($rel, 0, length($rel) - length("/smp_affinity")); if (system("mkdir -p $tmpdir/$dir")) { exit 1; } if (0) { print "cp $file $tmpdir/$rel\n"; } if (system("sudo cat $file > $tmpdir/$rel")) { exit 1; } }'; then
|
||||
echo "error: saving /proc/irq/*/smp_affinity" >&2
|
||||
error_exit "irqbalance_stopped"
|
||||
fi;
|
||||
@@ -245,7 +255,7 @@ if [ "${irqbalance_used}" == "yes" ]; then
|
||||
smp_affinity_mask=`echo $cpus | ncpus=$ncpus perl -e 'while(<>){@tokens = split /,/;foreach $token (@tokens) {@nums = split /-/,$token; for($num = $nums[0]; $num <= $nums[$#nums]; $num++) {$ndx=int($num/32); $mask[$ndx] |= (1<<($num % 32))}}} $nint32s = int(($ENV{'ncpus'}+31)/32); for($j = $nint32s - 1; $j >= 0; $j--) { if($j != $nint32s - 1){print ",";} $nblks = ($j != $nint32s - 1) ? 8 : ($ENV{'ncpus'} % 32 != 0) ? int((($ENV{'ncpus'} + 3) % 32) / 4) : 8; for($i = $nblks - 1;$i >= 0;$i--){ printf("%01x",($mask[$j] >> ($i*4)) & 0xf);}}'`
|
||||
# echo cpus=$cpus ncpus=$ncpus smp_affinity_mask=$smp_affinity_mask
|
||||
|
||||
if ! ncpus=$ncpus smp_affinity_mask=$smp_affinity_mask perl -e '@dirs = grep { -d } glob "/proc/irq/*"; foreach $dir (@dirs) { $hit = 0; $affinity_str = `cat $dir/smp_affinity`; chomp $affinity_str; @int32strs = split /,/, $affinity_str; @int32strs_mask=split /,/, $ENV{'smp_affinity_mask'}; for($i=0;$i <= $#int32strs_mask; $i++) { $int32strs_inv[$i] = sprintf("%08x",hex($int32strs_mask[$i])^0xffffffff); if($i == 0) { $len = int((($ENV{'ncpus'}%32)+3)/4); if($len != 0) { $int32strs_inv[$i] = substr($int32strs_inv[$i], -$len, $len); } } } $inv = join(",", @int32strs_inv); $nint32s = int(($ENV{'ncpus'}+31)/32); for($j = $nint32s - 1; $j >= 0; $j--) { if(hex($int32strs[$nint32s - 1 - $j]) & hex($int32strs_mask[$nint32s - 1 - $j])) { $hit = 1; }} if($hit == 1) { $cmd = "echo $inv > $dir/smp_affinity 2>/dev/null"; system $cmd;}}'; then
|
||||
if ! ncpus=$ncpus smp_affinity_mask=$smp_affinity_mask perl -e '@dirs = grep { -d } glob "/proc/irq/*"; foreach $dir (@dirs) { $hit = 0; $affinity_str = `sudo cat $dir/smp_affinity`; chomp $affinity_str; @int32strs = split /,/, $affinity_str; @int32strs_mask=split /,/, $ENV{'smp_affinity_mask'}; for($i=0;$i <= $#int32strs_mask; $i++) { $int32strs_inv[$i] = sprintf("%08x",hex($int32strs_mask[$i])^0xffffffff); if($i == 0) { $len = int((($ENV{'ncpus'}%32)+3)/4); if($len != 0) { $int32strs_inv[$i] = substr($int32strs_inv[$i], -$len, $len); } } } $inv = join(",", @int32strs_inv); $nint32s = int(($ENV{'ncpus'}+31)/32); for($j = $nint32s - 1; $j >= 0; $j--) { if(hex($int32strs[$nint32s - 1 - $j]) & hex($int32strs_mask[$nint32s - 1 - $j])) { $hit = 1; }} if($hit == 1) { system("echo $inv > /tmp/mcreboot/smp_affinity"); $cmd = "cp /tmp/mcreboot/smp_affinity $dir/smp_affinity 2>/dev/null"; system("sudo $cmd")}}'; then
|
||||
echo "error: modifying /proc/irq/*/smp_affinity" >&2
|
||||
error_exit "irqbalance_stopped"
|
||||
fi
|
||||
@@ -258,22 +268,22 @@ umask 0`printf "%o" ${umask_dec}`
|
||||
|
||||
# Load IHK if not loaded
|
||||
if ! grep -E 'ihk\s' /proc/modules &>/dev/null; then
|
||||
if ! taskset -c 0 insmod ${KMODDIR}/ihk.ko 2>/dev/null; then
|
||||
if ! taskset -c 0 sudo insmod ${KMODDIR}/ihk.ko 2>/dev/null; then
|
||||
echo "error: loading ihk" >&2
|
||||
error_exit "smp_affinity_modified"
|
||||
fi
|
||||
fi
|
||||
|
||||
# Increase swappiness so that we have better chance to allocate memory for IHK
|
||||
echo 100 > /proc/sys/vm/swappiness
|
||||
echo 100 | sudo tee /proc/sys/vm/swappiness 1>/dev/null
|
||||
|
||||
# Drop Linux caches to free memory
|
||||
sync && echo 3 > /proc/sys/vm/drop_caches
|
||||
sudo sync && echo 3 | sudo tee /proc/sys/vm/drop_caches 1>/dev/null
|
||||
|
||||
# Merge free memory areas into large, physically contigous ones
|
||||
echo 1 > /proc/sys/vm/compact_memory 2>/dev/null
|
||||
echo 1 | sudo tee /proc/sys/vm/compact_memory 1>/dev/null 2>/dev/null
|
||||
|
||||
sync
|
||||
sudo sync
|
||||
|
||||
# Load IHK-SMP if not loaded and reserve CPUs and memory
|
||||
if ! grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then
|
||||
@@ -289,7 +299,7 @@ if ! grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then
|
||||
error_exit "ihk_loaded"
|
||||
fi
|
||||
fi
|
||||
if ! taskset -c 0 insmod ${KMODDIR}/ihk-smp-@ARCH@.ko ihk_start_irq=$ihk_irq ihk_ikc_irq_core=$ihk_ikc_irq_core 2>/dev/null; then
|
||||
if ! taskset -c 0 sudo insmod ${KMODDIR}/ihk-smp-@ARCH@.ko ihk_start_irq=$ihk_irq ihk_ikc_irq_core=$ihk_ikc_irq_core 2>/dev/null; then
|
||||
echo "error: loading ihk-smp-@ARCH@" >&2
|
||||
error_exit "ihk_loaded"
|
||||
fi
|
||||
@@ -298,20 +308,20 @@ if ! grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then
|
||||
if [ "`hostname | grep "c[0-9][0-9][0-9][0-9].ofp"`" != "" ] && [ "`cat /sys/devices/system/node/online`" == "0-7" ]; then
|
||||
for i in 0 1 2 3; do
|
||||
find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
|
||||
echo 0 > $f 2>&1 > /dev/null;
|
||||
echo 0 | sudo tee $f 2>/dev/null
|
||||
done
|
||||
find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
|
||||
echo 1 > $f 2>&1 > /dev/null;
|
||||
echo 1 | sudo tee $f 2>/dev/null
|
||||
done
|
||||
done
|
||||
for i in 4 5 6 7; do
|
||||
find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
|
||||
echo 0 > $f 2>&1 > /dev/null;
|
||||
echo 0 | sudo tee $f 2>/dev/null
|
||||
done
|
||||
done
|
||||
for i in 4 5 6 7; do
|
||||
find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
|
||||
echo 1 > $f 2>&1 > /dev/null;
|
||||
echo 1 | sudo tee $f 2>/dev/null
|
||||
done
|
||||
done
|
||||
fi
|
||||
@@ -320,21 +330,21 @@ if ! grep ihk_smp_@ARCH@ /proc/modules &>/dev/null; then
|
||||
if [ "`hostname | grep "c[0-9][0-9][0-9][0-9].ofp"`" != "" ] && [ "`cat /sys/devices/system/node/online`" == "0-1" ]; then
|
||||
for i in 1; do
|
||||
find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
|
||||
echo 0 > $f 2>&1 > /dev/null;
|
||||
echo 0 | sudo tee $f 2>/dev/null
|
||||
done
|
||||
done
|
||||
for i in 1; do
|
||||
find /sys/devices/system/node/node$i/memory*/ -name "online" | while read f; do
|
||||
echo 1 > $f 2>&1 > /dev/null;
|
||||
echo 1 | sudo tee $f 2>/dev/null
|
||||
done
|
||||
done
|
||||
fi
|
||||
|
||||
if ! ${SBINDIR}/ihkconfig 0 reserve mem ${mem}; then
|
||||
if ! sudo ${SBINDIR}/ihkconfig 0 reserve mem ${mem}; then
|
||||
echo "error: reserving memory" >&2
|
||||
error_exit "ihk_smp_loaded"
|
||||
fi
|
||||
if ! ${SBINDIR}/ihkconfig 0 reserve cpu ${cpus}; then
|
||||
if ! sudo ${SBINDIR}/ihkconfig 0 reserve cpu ${cpus}; then
|
||||
echo "error: reserving CPUs" >&2;
|
||||
error_exit "mem_reserved"
|
||||
fi
|
||||
@@ -342,14 +352,14 @@ fi
|
||||
|
||||
# Load mcctrl if not loaded
|
||||
if ! grep mcctrl /proc/modules &>/dev/null; then
|
||||
if ! taskset -c 0 insmod ${KMODDIR}/mcctrl.ko 2>/dev/null; then
|
||||
if ! taskset -c 0 sudo insmod ${KMODDIR}/mcctrl.ko 2>/dev/null; then
|
||||
echo "error: inserting mcctrl.ko" >&2
|
||||
error_exit "cpus_reserved"
|
||||
fi
|
||||
fi
|
||||
|
||||
# Check that different versions of binaries/scripts are not mixed
|
||||
IHK_BUILDID=`${SBINDIR}/ihkconfig 0 get buildid`
|
||||
IHK_BUILDID=`sudo ${SBINDIR}/ihkconfig 0 get buildid`
|
||||
if [ "${IHK_BUILDID}" != "${MCK_BUILDID}" ]; then
|
||||
echo "IHK build-id (${IHK_BUILDID}) didn't match McKernel build-id (${MCK_BUILDID})." >&2
|
||||
error_exit "mcctrl_loaded"
|
||||
@@ -361,7 +371,7 @@ if ls /dev/mcos* 1>/dev/null 2>&1; then
|
||||
ind=`echo $i|cut -c10-`;
|
||||
# Retry when conflicting with ihkmond
|
||||
nretry=0
|
||||
until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -ge 4 ]; do
|
||||
until sudo ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -ge 4 ]; do
|
||||
sleep 0.25
|
||||
nretry=$[ $nretry + 1 ]
|
||||
done
|
||||
@@ -373,51 +383,51 @@ if ls /dev/mcos* 1>/dev/null 2>&1; then
|
||||
fi
|
||||
|
||||
# Create OS instance
|
||||
if ! ${SBINDIR}/ihkconfig 0 create; then
|
||||
if ! sudo ${SBINDIR}/ihkconfig 0 create; then
|
||||
echo "error: creating OS instance" >&2
|
||||
error_exit "mcctrl_loaded"
|
||||
fi
|
||||
|
||||
# Assign CPUs
|
||||
if ! ${SBINDIR}/ihkosctl 0 assign cpu ${cpus}; then
|
||||
if ! sudo ${SBINDIR}/ihkosctl 0 assign cpu ${cpus}; then
|
||||
echo "error: assign CPUs" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
if [ "$ikc_map" != "" ]; then
|
||||
# Specify IKC map
|
||||
if ! ${SBINDIR}/ihkosctl 0 set ikc_map ${ikc_map}; then
|
||||
if ! sudo ${SBINDIR}/ihkosctl 0 set ikc_map ${ikc_map}; then
|
||||
echo "error: assign CPUs" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
fi
|
||||
|
||||
# Assign memory
|
||||
if ! ${SBINDIR}/ihkosctl 0 assign mem ${mem}; then
|
||||
if ! sudo ${SBINDIR}/ihkosctl 0 assign mem ${mem}; then
|
||||
echo "error: assign memory" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
# Load kernel image
|
||||
if ! ${SBINDIR}/ihkosctl 0 load ${KERNDIR}/mckernel.img; then
|
||||
if ! sudo ${SBINDIR}/ihkosctl 0 load ${KERNDIR}/mckernel.img; then
|
||||
echo "error: loading kernel image: ${KERNDIR}/mckernel.img" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
# Set kernel arguments
|
||||
if ! ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo $safe_kernel_map $idle_halt dump_level=${DUMP_LEVEL} $extra_kopts $allow_oversubscribe $time_sharing"; then
|
||||
if ! sudo ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo $safe_kernel_map $idle_halt dump_level=${DUMP_LEVEL} $extra_kopts $allow_oversubscribe $time_sharing"; then
|
||||
echo "error: setting kernel arguments" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
# Boot OS instance
|
||||
if ! ${SBINDIR}/ihkosctl 0 boot; then
|
||||
if ! sudo ${SBINDIR}/ihkosctl 0 boot; then
|
||||
echo "error: booting" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
# Set device file ownership
|
||||
if ! chown ${chown_option} /dev/mcd* /dev/mcos*; then
|
||||
if ! sudo chown ${chown_option} /dev/mcd* /dev/mcos*; then
|
||||
echo "warning: failed to chown device files" >&2
|
||||
fi
|
||||
|
||||
@@ -429,12 +439,12 @@ if [ "${irqbalance_used}" == "yes" ]; then
|
||||
|
||||
disable_irqbalance_mck
|
||||
|
||||
if ! systemctl link $ETCDIR/irqbalance_mck.service >/dev/null 2>/dev/null; then
|
||||
if ! sudo systemctl link $ETCDIR/irqbalance_mck.service >/dev/null 2>/dev/null; then
|
||||
echo "error: linking irqbalance_mck" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
if ! systemctl start irqbalance_mck.service 2>/dev/null ; then
|
||||
if ! sudo systemctl start irqbalance_mck.service 2>/dev/null ; then
|
||||
echo "error: starting irqbalance_mck" >&2
|
||||
error_exit "os_created"
|
||||
fi
|
||||
|
||||
Reference in New Issue
Block a user