#!/bin/bash


OWNNAME=`basename $0`

if [ "$1" = "-h" ]; then
	cat <<-EOF
		Usage: $OWNNAME [ -n | -p | -s ] | -t | -v
		Show informations about processes with huge memory pages allocated.

		Without giving any option, $OWNNAME will output a process-oriented list
		sorted by PID.

		  Options for process-oriented listing:
		  -n	sort by NUMA node
		  -p	sort by page size
		  -s	sort by size

		  -t	CPU-topology oriented output, sorted by NUMA-node and core

		  -v	output version information and exit
		  -h	this cruft.

		Any processname starting with "qemu-system" will be replaced by the username of
		the process.

		$OWNNAME assumes that all huge pages of a process are allocated from a
		single NUMA-node. Processes with huge pages spread over more than one node may
		cause $OWNNAME to produce errors or even incorrect results.
	EOF
	exit
fi

if [ "$1" = "-v" ]; then
	cat <<- EOF

	  This is lshugemem.sh rev.0.1 by Kadir Mueller (kadir.mueller@theflatnet.de),
	  distributed in 2018 under the terms of version 2 of the GNU public license.

	  There may be updates avalable at: http://www.theflatnet.de/pub/stuff/numa/
	  Any feedback according this script is highly appreciated.

	EOF
	exit
fi

PIDS=`grep -w huge /proc/[0-9]*/numa_maps 2>/dev/null | cut -d / -f 3 | sort | uniq`
declare -a PROCS
declare -a PAGESIZES
declare -a PAGES
declare -a SIZES
declare -a CPUS
declare -a NODES
declare -a CORES
declare -a CPUNPROCS
declare -a CPUNPROCNAMES

MAXPROCNAMELEN=0
NCPUS=0
while [ "$NCPUS" != `ls -d /sys/devices/system/cpu/cpu[0-9]* | wc -l` ]; do
	CPUNPROCS[$NCPUS]=0
	let NCPUS=$NCPUS+1
done
for PID in $PIDS; do
	PROCS[$PID]=`ps -edaf | grep -vw grep | awk '$2=='$PID' { print $8 }'`
	PAGESIZES[$PID]="`grep -w huge /proc/$PID/numa_maps | awk -F= '{ print $NF }' | sort | uniq`"
	if [ "${PAGESIZES[$PID]}" != "" ]; then
		echo ${PROCS[$PID]} | grep -q ^qemu-system && PROCS[$PID]="`ls -ld /proc/$PID | awk '{ print $3 }'`"
		[ `echo -n ${PROCS[$PID]} | wc -c` -gt $MAXPROCNAMELEN ] && MAXPROCNAMELEN=`echo -n ${PROCS[$PID]} | wc -c`
		PAGES[$PID]=`grep -w huge /proc/$PID/numa_maps | tr ' ' '\n' | grep ^N[0-9]\*= | cut -d = -f 2 | tr '\n' + | sed 's/+$//'`
		PAGES[$PID]=`echo ${PAGES[$PID]} | bc`
		let PAGESIZES[$PID]=${PAGESIZES[$PID]}/1024
		let SIZES[$PID]=${PAGES[$PID]}\*${PAGESIZES[$PID]}
		CPUS[$PID]=`taskset -cp $PID | awk '{ print $NF }' | tr , ' '`
		for CPU in ${CPUS[$PID]}; do
			let CPUNPROCS[$CPU]=${CPUNPROCS[$CPU]}+1
			CPUNPROCNAMES[$CPU]="${CPUNPROCNAMES[$CPU]} ${PROCS[$PID]}"
		done
		NODES[$PID]=`for CPU in ${CPUS[$PID]}; do
			cat /sys/devices/system/cpu/cpu$CPU/topology/physical_package_id
		done | sort | uniq | tr '\n' ' '| sed 's/ $//'`
		CORES[$PID]=`for CPU in ${CPUS[$PID]}; do
			cat /sys/devices/system/cpu/cpu$CPU/topology/core_id
		done | sort | uniq | tr '\n' ' ' | sed 's/ $//'`
	fi
done

if [ "$1" != "-t" ]; then
	printf "   PID %${MAXPROCNAMELEN}s   Size PgSzM Pges    On CPU(s) Core(s) Node(s)\n" Name
	printf " ----- %${MAXPROCNAMELEN}s ------ ----- ---- ------------ ------- -------\n" "`printf "%${MAXPROCNAMELEN}s" " " | tr ' ' -`"
	for PID in $PIDS; do
		SORT=$PID
		[ "$1" = "-n" ] && SORT="${NODES[$PID]}"
		[ "$1" = "-s" ] && SORT="${SIZES[$PID]}"
		[ "$1" = "-p" ] && SORT="${PAGESIZES[$PID]}"
		printf "%20s %6i %${MAXPROCNAMELEN}s %6i %5i %4i %12s %7s %6s\n" $SORT $PID "${PROCS[$PID]}" ${SIZES[$PID]} ${PAGESIZES[$PID]} ${PAGES[$PID]} "${CPUS[$PID]}" "${CORES[$PID]}" "${NODES[$PID]}"
	done | sort -n | cut -c 22-
else
	echo "Node Core  CPU Cnt Names"
	for CPUPATH in `ls -d /sys/devices/system/node/node[0-9]*/cpu[0-9]*`; do
		CPU=`echo $CPUPATH | awk -F/ '{ print $NF }' | sed 's/cpu//'`
		NODE=`echo $CPUPATH | awk -F/ '{ print $6 }' | sed 's/node//'`
		printf "%.3i%.3i%.3i%4i %4i %4i %3i%s\n" $NODE `cat /sys/devices/system/cpu/cpu$CPU/topology/core_id` $CPU $NODE `cat /sys/devices/system/cpu/cpu$CPU/topology/core_id` $CPU ${CPUNPROCS[$CPU]} "${CPUNPROCNAMES[$CPU]}"
	done | sort -n | cut -c 10-
fi