gh0s7/mckernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Redirect kmsg to /dev/log and detect hungup

Browse Source 
1. ihkmond retrieves kmsg when the amount of kmsg exceeds the threashold and
   /dev/mcosX is deleted
2. ihkmond periodically monitors OS status change to detect hungup
This commit is contained in:
Masamichi Takagi
2017-09-01 15:31:37 +09:00
parent daa7526127
commit 5b51eb80a3
7 changed files with 138 additions and 186 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
arch/x86/tools/mcreboot-smp-x86.sh.in
View File

@ -29,8 +29,8 @@ if [ "${BASH_VERSINFO[0]}" -lt 4 ]; then
exit 1
fi
INTERVAL=1
LOGMODE=0
redirect_kmsg=0
mon_interval="-1"
DUMP_LEVEL=24
facility="LOG_LOCAL6"
chown_option=`logname 2> /dev/null`
@ -44,38 +44,14 @@ fi
turbo=""
ihk_irq=""
while getopts :ti:k:c:m:o:f:r:q:d: OPT
while getopts :tk:c:m:o:f:r:q:i:d: OPT
do
case ${OPT} in
f) facility=${OPTARG}
;;
o) chown_option=${OPTARG}
;;
i) INTERVAL=${OPTARG}
expr "${INTERVAL}" + 1 > /dev/null 2>&1
if [ $? -ge 2 ]
then
echo "invalid -i value" >&2
exit 1
fi
if [ ${INTERVAL} -le 0 ]
then
echo "invalid -i value" >&2
exit 1
fi
;;
k) LOGMODE=${OPTARG}
expr "${LOGMODE}" + 1 > /dev/null 2>&1
if [ $? -ge 2 ]
then
echo "invalid -k value" >&2
exit 1
fi
if [ ${LOGMODE} -lt 0 -o ${LOGMODE} -gt 2 ]
then
echo "invalid -k value" >&2
exit 1
fi
k) redirect_kmsg=${OPTARG}
;;
c) cpus=${OPTARG}
;;
@ -89,11 +65,21 @@ do
;;
d) DUMP_LEVEL=${OPTARG}
;;
i) mon_interval=${OPTARG}
;;
*) echo "invalid option -${OPT}" >&2
exit 1
esac
done
# Start ihkmond
pid=`pidof ihkmond`
if [ "${pid}" != "" ]; then
sudo kill -9 ${pid} > /dev/null 2> /dev/null
fi
if [ "${redirect_kmsg}" != "0" -o "${mon_interval}" != "-1" ]; then
${SBINDIR}/ihkmond -f ${facility} -k ${redirect_kmsg} -i ${mon_interval}
fi
#
# Revert any state that has been initialized before the error occured.
#
@ -248,7 +234,7 @@ if [ "${irqbalance_used}" == "yes" ]; then
if ! systemctl stop irqbalance.service 2>/dev/null ; then
echo "error: stopping irqbalance" >&2
exit 1
fi;
fi;
if ! etcdir=@ETCDIR@ perl -e 'use File::Copy qw(copy); $etcdir=$ENV{'etcdir'}; @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(0) { print "cp $file $etcdir/$rel\n";} if(system("mkdir -p $etcdir/$dir")){ exit 1;} if(!copy($file,"$etcdir/$rel")){ exit 1;} }'; then
echo "error: saving /proc/irq/*/smp_affinity" >&2
@ -351,9 +337,15 @@ fi
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
echo "error: destroying LWK instance $ind failed" >&2
error_exit "mcctrl_loaded"
# Retry when conflicting with ihkmond
nretry=0
until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -lt 4 ]; do
sleep 0.25
nretry=$[ $nretry + 1 ]
done
if [ $nretry -eq 4 ]; then
echo "error: destroying LWK instance $ind failed" >&2
error_exit "mcctrl_loaded"
fi
done
fi
@ -391,7 +383,7 @@ if ! ${SBINDIR}/ihkosctl 0 load ${KERNDIR}/mckernel.img; then
fi
# Set kernel arguments
if ! ${SBINDIR}/ihkosctl 0 kargs "hidos ksyslogd=${LOGMODE} $turbo dump_level=${DUMP_LEVEL}"; then
if ! ${SBINDIR}/ihkosctl 0 kargs "hidos $turbo dump_level=${DUMP_LEVEL}"; then
echo "error: setting kernel arguments" >&2
error_exit "os_created"
fi
@ -429,14 +421,3 @@ if [ "${irqbalance_used}" == "yes" ]; then
fi
# echo cpus=$cpus ncpus=$ncpus banirq=$banirq
fi
# Start mcklogd. Note that McKernel blocks when kmsg buffer is full
# with '-k 1' until mcklogd unblocks it so starting mcklogd must preceed
# booting McKernel
if [ ${LOGMODE} -ne 0 ]; then
# Stop mcklogd which has survived McKernel shutdown because
# mcstop+release.sh is not used
pkill mcklogd
SBINDIR=${SBINDIR} ${SBINDIR}/mcklogd -i ${INTERVAL} -f ${facility}
fi

24
arch/x86/tools/mcstop+release-smp-x86.sh.in
View File

@ -21,12 +21,6 @@ irqbalance_used=""
# No SMP module? Exit.
if ! grep ihk_smp_x86 /proc/modules &>/dev/null; then exit 0; fi
# Stop mcklogd
while pgrep "mcklogd" > /dev/null 2>&1;
do
pkill -9 mcklogd
done
if [ "`systemctl status irqbalance_mck.service 2> /dev/null |grep -E 'Active: active'`" != "" ]; then
irqbalance_used="yes"
if ! systemctl stop irqbalance_mck.service 2>/dev/null; then
@ -41,9 +35,15 @@ fi
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
echo "error: destroying LWK instance $ind failed" >&2
exit 1
# Retry when conflicting with ihkmond
nretry=0
until ${SBINDIR}/ihkconfig 0 destroy $ind || [ $nretry -lt 4 ]; do
sleep 0.25
nretry=$[ $nretry + 1 ]
done
if [ $nretry -eq 4 ]; then
echo "error: destroying LWK instance $ind failed" >&2
exit 1
fi
done
fi
@ -102,6 +102,12 @@ if grep -E 'ihk\s' /proc/modules &>/dev/null; then
fi
fi
# Stop ihkmond
pid=`pidof ihkmond`
if [ "${pid}" != "" ]; then
sudo kill -9 ${pid} > /dev/null 2> /dev/null
fi
# Start irqbalance with the original settings
if [ "${irqbalance_used}" != "" ]; then
if ! etcdir=@ETCDIR@ perl -e '$etcdir=$ENV{'etcdir'}; @files = grep { -f } glob "$etcdir/proc/irq/*/smp_affinity"; foreach $file (@files) { $dest = substr($file, length($etcdir)); if(0) {print "cp $file $dest\n";} system("cp $file $dest 2>/dev/null"); }'; then