resstalls

#!/bin/bash
#
# resstalls - measure and summarize resource stalls.
#            Uses Linux perf and PMCs.
#
# FROM: https://github.com/brendangregg/pmc-cloud-tools
#
# USAGE: resstalls {-C CPU | -p PID | -c CMD} [interval [duration]]
#
# Columns:
#
# - ALL: Resource stalls x 1M
# - LOAD: Load buffer stalls x 1M
# - PIPE: Instruction pipeline reservation station full stalls x 1M
# - STORE: Store related stall x 1M
# - ROB: Re-order buffer full store x 1M
# - FLOAT: Floating point unit control word stall x 1M
# - MXCSR: MXCSR register rename stall x 1M
# - OTHER: Other stall reasons x 1M
# - SUM: Sum of all columns except ALL
#
# Copyright 2020 Netflix, Inc.
# Licensed under the Apache License, Version 2.0 (the "License")
#
# 24-Jan-2020	Brendan Gregg	Created this.

function usage {
	cat <<-END >&2
	USAGE: resstalls {-C CPU | -p PID | -c CMD} [interval [duration]]
	                 -C CPU         # measure this CPU only
	                 -p PID         # measure this PID only
	                 -c 'CMD'       # measure this command only (quote it)
	                 interval       # output interval in secs (default 1)
	                 duration       # total seconds (default infinityish)
	  eg,
	       resstalls                 # show stats across all CPUs
	       resstalls 5               # show stats every 5 seconds
	       resstalls -C 0            # measure CPU 0 only
	       resstalls -p 181          # measure PID 181 only
	       resstalls -c 'cksum /boot/*' # measure run and measure this cmd
END
	exit
}

opt_cpu=0; opt_pid=0; opt_cmd=0; cpu=""; pid=""; cmd=""

while getopts C:p:c:h opt
do
	case $opt in
	C)	opt_cpu=1; cpu=$OPTARG ;;
	p)	opt_pid=1; pid=$OPTARG ;;
	c)	opt_cmd=1; cmd=$OPTARG ;;
	h|?)	usage ;;
	esac
done
shift $(( $OPTIND - 1 ))

if (( opt_cpu + opt_pid + opt_cmd > 1 )); then
	echo >&2 "ERROR: pick one of -C, -p, -c"
	usage
fi
secs=${1:-1}			# default 1 second
duration=${2:-999999999}	# default semi-infinite seconds
hlines=25			# lines to repeat header
target=-a
(( opt_cpu )) && target="-C $cpu sleep $duration"
(( opt_pid )) && target="-p $pid sleep $duration"
(( opt_cmd )) && target="$cmd"

if (( opt_pid )); then
	if [ ! -d /proc/$pid ]; then
		echo >&2 "ERROR: Can't find PID $pid. Exiting."
		exit
	fi
fi

# note that instructions is last on purpose, it triggers output
# cycles are twice as a workaround for an issue
# the r4f2e and r412e counters are from the architectural set, so should be stable
echo "All counter columns are x 1000000"
perf stat -e resource_stalls.any,r02a2,r04a2,r08a2,r10a2,r20a2,r40a2,r80a2 \
	-I $(( secs * 1000 )) $target 2>&1 | awk -v hlines=$hlines '
	BEGIN {
		htxt = sprintf("%8s %8s %8s %8s %8s %8s %8s %8s %8s",
			"ALL", "LOAD", "PIPE", "STORE", "ROB", "FLOAT",
			"MXCSR", "OTHER", "SUM");
		print htxt
		header = hlines
	}
	/invalid/ { print $0 }	# unsupported event
	{ gsub(/,/, ""); }
	$3 == "resource_stalls.any" { rsany = $2 }	# RESOURCE_STALLS.ANY
	$3 == "r02a2" { rsload = $2 }			# RESOURCE_STALLS.LOAD
	$3 == "r04a2" { rspipe = $2 }			# RESOURCE_STALLS.RS_FULL
	$3 == "r08a2" { rsstore = $2 }			# RESOURCE_STALLS.STORE
	$3 == "r10a2" { rsrob = $2 }			# RESOURCE_STALLS.ROB_FULL
	$3 == "r20a2" { rsfpcw = $2 }			# RESOURCE_STALLS.FPCW
	$3 == "r40a2" { rsmxcsr = $2 }			# RESOURCE_STALLS.MXCSR
	$3 == "r80a2" {				# last one, trigger output
		rsother = $2				# RESOURCE_STALLS.OTHER
		if (--header == 0) {
			print htxt
			header = hlines
		}

		rssum = rsload + rspipe + rsstore + rsrob + rsfpcw + rsmxcsr + rsother

		printf("%8.1f %8.1f %8.1f %8.1f %8.1f %8.1f %8.1f %8.1f %8.1f\n",
			rsany / 1000000, rsload / 1000000, rspipe / 1000000,
			rsstore / 1000000, rsrob / 1000000, rsfpcw / 1000000,
			rsmxcsr / 1000000, rsother / 1000000, rssum / 1000000)
	}
'