| 1 | #!/bin/bash |
| 2 | |
| 3 | args=$* |
| 4 | first_cores="" |
| 5 | taskset_cores="" |
| 6 | first_cores_count=0 |
| 7 | nb_threads=1 |
| 8 | drives="" |
| 9 | |
| 10 | # Default options |
| 11 | latency_cmdline="" |
| 12 | |
| 13 | fatal() { |
| 14 | echo "$@" |
| 15 | exit 1 |
| 16 | } |
| 17 | |
| 18 | hint() { |
| 19 | echo "Warning: $*" |
| 20 | } |
| 21 | |
| 22 | info() { |
| 23 | item=$1 |
| 24 | shift |
| 25 | echo "${item}: $*" |
| 26 | } |
| 27 | |
| 28 | check_root() { |
| 29 | [[ ${EUID} -eq 0 ]] || fatal "You should be root to run this tool" |
| 30 | } |
| 31 | |
| 32 | check_binary() { |
| 33 | # Ensure the binaries are present and executable |
| 34 | for bin in "$@"; do |
| 35 | if [ ! -x ${bin} ]; then |
| 36 | which ${bin} >/dev/null |
| 37 | [ $? -eq 0 ] || fatal "${bin} doesn't exists or is not executable" |
| 38 | fi |
| 39 | done |
| 40 | } |
| 41 | |
| 42 | detect_first_core() { |
| 43 | cpu_to_search="0" |
| 44 | if [ "${#drives[@]}" -eq 1 ]; then |
| 45 | device_name=$(block_dev_name ${drives[0]}) |
| 46 | device_dir="/sys/block/${device_name}/device/" |
| 47 | pci_addr=$(cat ${device_dir}/address) |
| 48 | pci_dir="/sys/bus/pci/devices/${pci_addr}/" |
| 49 | cpu_to_search=$(cat ${pci_dir}/local_cpulist | cut -d"," -f 1 | cut -d"-" -f 1) |
| 50 | else |
| 51 | hint 'Passed multiple devices. Running on the first core.' |
| 52 | fi |
| 53 | core_to_run=$(lscpu --all -pSOCKET,CORE,CPU | grep ",$cpu_to_search\$" | cut -d"," -f1-2) |
| 54 | |
| 55 | # Detect which logical cpus belongs to the first physical core |
| 56 | # If Hyperthreading is enabled, two cores are returned |
| 57 | cpus=$(lscpu --all -pSOCKET,CORE,CPU | grep "$core_to_run") |
| 58 | for cpu in ${cpus}; do |
| 59 | IFS=',' |
| 60 | # shellcheck disable=SC2206 |
| 61 | array=(${cpu}) |
| 62 | if [ ${first_cores_count} -eq 0 ]; then |
| 63 | first_cores="${array[2]}" |
| 64 | else |
| 65 | first_cores="${first_cores} ${array[2]}" |
| 66 | fi |
| 67 | |
| 68 | first_cores_count=$((first_cores_count + 1)) |
| 69 | unset IFS |
| 70 | done |
| 71 | [ ${first_cores_count} -eq 0 ] && fatal "Cannot detect first core" |
| 72 | taskset_cores=$(echo "${first_cores}" | tr ' ' ',') |
| 73 | } |
| 74 | |
| 75 | usage() { |
| 76 | echo "usage: [options] block_device [other_block_devices] |
| 77 | |
| 78 | -h : print help |
| 79 | -l : enable latency reporting |
| 80 | |
| 81 | example: |
| 82 | t/one-core-peak.sh /dev/nvme0n1 |
| 83 | t/one-core-peak.sh -l /dev/nvme0n1 /dev/nvme1n1 |
| 84 | " |
| 85 | exit 0 |
| 86 | } |
| 87 | |
| 88 | check_args() { |
| 89 | local OPTIND option |
| 90 | while getopts "hl" option; do |
| 91 | case "${option}" in |
| 92 | h) # Show help |
| 93 | usage |
| 94 | ;; |
| 95 | l) # Report latency |
| 96 | latency_cmdline="1" |
| 97 | ;; |
| 98 | *) |
| 99 | fatal "Unsupported ${option} option" |
| 100 | ;; |
| 101 | esac |
| 102 | done |
| 103 | shift $((OPTIND-1)) |
| 104 | [ $# -eq 0 ] && fatal "Missing drive(s) as argument" |
| 105 | drives="$*" |
| 106 | } |
| 107 | |
| 108 | check_drive_exists() { |
| 109 | # Ensure the block device exists |
| 110 | [ -b $1 ] || fatal "$1 is not a valid block device" |
| 111 | } |
| 112 | |
| 113 | is_nvme() { |
| 114 | [[ ${*} == *"nvme"* ]] |
| 115 | } |
| 116 | |
| 117 | check_poll_queue() { |
| 118 | # Print a warning if the nvme poll queues aren't enabled |
| 119 | is_nvme ${drives} || return |
| 120 | poll_queue=$(cat /sys/module/nvme/parameters/poll_queues) |
| 121 | [ ${poll_queue} -eq 0 ] && hint "For better performance, you should enable nvme poll queues by setting nvme.poll_queues=32 on the kernel commande line" |
| 122 | } |
| 123 | |
| 124 | block_dev_name() { |
| 125 | echo ${1#"/dev/"} |
| 126 | } |
| 127 | |
| 128 | get_sys_block_dir() { |
| 129 | # Returns the /sys/block/ directory of a given block device |
| 130 | device_name=$1 |
| 131 | sys_block_dir="/sys/block/${device_name}" |
| 132 | [ -d "${sys_block_dir}" ] || fatal "Cannot find ${sys_block_dir} directory" |
| 133 | echo ${sys_block_dir} |
| 134 | } |
| 135 | |
| 136 | check_io_scheduler() { |
| 137 | # Ensure io_sched is set to none |
| 138 | device_name=$(block_dev_name $1) |
| 139 | sys_block_dir=$(get_sys_block_dir ${device_name}) |
| 140 | sched_file="${sys_block_dir}/queue/scheduler" |
| 141 | [ -f "${sched_file}" ] || fatal "Cannot find IO scheduler for ${device_name}" |
| 142 | grep -q '\[none\]' ${sched_file} |
| 143 | if [ $? -ne 0 ]; then |
| 144 | info "${device_name}" "set none as io scheduler" |
| 145 | echo "none" > ${sched_file} |
| 146 | fi |
| 147 | |
| 148 | } |
| 149 | |
| 150 | check_sysblock_value() { |
| 151 | device_name=$(block_dev_name $1) |
| 152 | sys_block_dir=$(get_sys_block_dir ${device_name}) |
| 153 | target_file="${sys_block_dir}/$2" |
| 154 | value=$3 |
| 155 | [ -f "${target_file}" ] || return |
| 156 | content=$(cat ${target_file} 2>/dev/null) |
| 157 | if [ "${content}" != "${value}" ]; then |
| 158 | echo ${value} > ${target_file} 2>/dev/null && info "${device_name}" "${target_file} set to ${value}." || hint "${device_name}: Cannot set ${value} on ${target_file}" |
| 159 | fi |
| 160 | } |
| 161 | |
| 162 | compute_nb_threads() { |
| 163 | # Increase the number of threads if there is more devices or cores than the default value |
| 164 | [ $# -gt ${nb_threads} ] && nb_threads=$# |
| 165 | [ ${first_cores_count} -gt ${nb_threads} ] && nb_threads=${first_cores_count} |
| 166 | } |
| 167 | |
| 168 | check_scaling_governor() { |
| 169 | driver=$(LC_ALL=C cpupower frequency-info |grep "driver:" |awk '{print $2}') |
| 170 | if [ -z "${driver}" ]; then |
| 171 | hint "Cannot detect processor scaling driver" |
| 172 | return |
| 173 | fi |
| 174 | cpupower frequency-set -g performance >/dev/null 2>&1 || fatal "Cannot set scaling processor governor" |
| 175 | } |
| 176 | |
| 177 | check_idle_governor() { |
| 178 | filename="/sys/devices/system/cpu/cpuidle/current_governor" |
| 179 | if [ ! -f "${filename}" ]; then |
| 180 | hint "Cannot detect cpu idle governor" |
| 181 | return |
| 182 | fi |
| 183 | echo "menu" > ${filename} 2>/dev/null || fatal "Cannot set cpu idle governor to menu" |
| 184 | } |
| 185 | |
| 186 | show_nvme() { |
| 187 | device="$1" |
| 188 | device_name=$(block_dev_name $1) |
| 189 | device_dir="/sys/block/${device_name}/device/" |
| 190 | pci_addr=$(cat ${device_dir}/address) |
| 191 | pci_dir="/sys/bus/pci/devices/${pci_addr}/" |
| 192 | link_speed=$(cat ${pci_dir}/current_link_speed) |
| 193 | irq=$(cat ${pci_dir}/irq) |
| 194 | numa=$([ -f ${pci_dir}/numa_node ] && cat ${pci_dir}/numa_node || echo "off") |
| 195 | cpus=$(cat ${pci_dir}/local_cpulist) |
| 196 | model=$(cat ${device_dir}/model | xargs) #xargs for trimming spaces |
| 197 | fw=$(cat ${device_dir}/firmware_rev | xargs) #xargs for trimming spaces |
| 198 | serial=$(cat ${device_dir}/serial | xargs) #xargs for trimming spaces |
| 199 | info ${device_name} "MODEL=${model} FW=${fw} serial=${serial} PCI=${pci_addr}@${link_speed} IRQ=${irq} NUMA=${numa} CPUS=${cpus} " |
| 200 | which nvme &> /dev/null |
| 201 | if [ $? -eq 0 ]; then |
| 202 | status="" |
| 203 | NCQA=$(nvme get-feature -H -f 0x7 ${device} 2>&1 |grep NCQA |cut -d ':' -f 2 | xargs) |
| 204 | [ -n "${NCQA}" ] && status="${status}Completion Queues:${NCQA}, " |
| 205 | NSQA=$(nvme get-feature -H -f 0x7 ${device} 2>&1 |grep NSQA |cut -d ':' -f 2 | xargs) |
| 206 | [ -n "${NSQA}" ] && status="${status}Submission Queues:${NSQA}, " |
| 207 | power_state=$(nvme get-feature -H -f 0x2 ${device} 2>&1 | grep PS |cut -d ":" -f 2 | xargs) |
| 208 | [ -n "${power_state}" ] && status="${status}PowerState:${power_state}, " |
| 209 | apste=$(nvme get-feature -H -f 0xc ${device} 2>&1 | grep APSTE |cut -d ":" -f 2 | xargs) |
| 210 | [ -n "${apste}" ] && status="${status} Autonomous Power State Transition:${apste}, " |
| 211 | temp=$(nvme smart-log ${device} 2>&1 |grep 'temperature' |cut -d ':' -f 2 |xargs) |
| 212 | [ -n "${temp}" ] && status="${status}Temp:${temp}" |
| 213 | info ${device_name} "${status}" |
| 214 | fi |
| 215 | } |
| 216 | |
| 217 | show_device() { |
| 218 | device_name=$(block_dev_name $1) |
| 219 | is_nvme $1 && show_nvme $1 |
| 220 | } |
| 221 | |
| 222 | show_kernel_config_item() { |
| 223 | config_item="CONFIG_$1" |
| 224 | config_file="/boot/config-$(uname -r)" |
| 225 | if [ ! -f "${config_file}" ]; then |
| 226 | config_file='/proc/config.gz' |
| 227 | if [ ! -f "${config_file}" ]; then |
| 228 | return |
| 229 | fi |
| 230 | fi |
| 231 | status=$(zgrep ${config_item}= ${config_file}) |
| 232 | if [ -z "${status}" ]; then |
| 233 | echo "${config_item}=N" |
| 234 | else |
| 235 | echo "${config_item}=$(echo ${status} | cut -d '=' -f 2)" |
| 236 | fi |
| 237 | } |
| 238 | |
| 239 | show_system() { |
| 240 | CPU_MODEL=$(grep -m1 "model name" /proc/cpuinfo | awk '{print substr($0, index($0,$4))}') |
| 241 | MEMORY_SPEED=$(dmidecode -t 17 -q | grep -m 1 "Configured Memory Speed: [0-9]" | awk '{print substr($0, index($0,$4))}') |
| 242 | KERNEL=$(uname -r) |
| 243 | info "system" "CPU: ${CPU_MODEL}" |
| 244 | info "system" "MEMORY: ${MEMORY_SPEED}" |
| 245 | info "system" "KERNEL: ${KERNEL}" |
| 246 | for config_item in BLK_CGROUP BLK_WBT_MQ HZ RETPOLINE PAGE_TABLE_ISOLATION; do |
| 247 | info "system" "KERNEL: $(show_kernel_config_item ${config_item})" |
| 248 | done |
| 249 | info "system" "KERNEL: $(cat /proc/cmdline)" |
| 250 | info "system" "SElinux: $(getenforce)" |
| 251 | tsc=$(journalctl -k | grep 'tsc: Refined TSC clocksource calibration:' | awk '{print $11}') |
| 252 | if [ -n "${tsc}" ]; then |
| 253 | info "system" "TSC: ${tsc} Mhz" |
| 254 | tsc=$(echo ${tsc} | tr -d '.') |
| 255 | [ -n "${latency_cmdline}" ] && latency_cmdline="-t1 -T${tsc}000" |
| 256 | fi |
| 257 | } |
| 258 | |
| 259 | ### MAIN |
| 260 | check_args ${args} |
| 261 | check_root |
| 262 | check_binary t/io_uring lscpu grep taskset cpupower awk tr xargs dmidecode |
| 263 | detect_first_core |
| 264 | |
| 265 | info "##################################################" |
| 266 | show_system |
| 267 | for drive in ${drives}; do |
| 268 | check_drive_exists ${drive} |
| 269 | check_io_scheduler ${drive} |
| 270 | check_sysblock_value ${drive} "queue/iostats" 0 # Ensure iostats are disabled |
| 271 | check_sysblock_value ${drive} "queue/nomerges" 2 # Ensure merge are disabled |
| 272 | check_sysblock_value ${drive} "queue/io_poll" 1 # Ensure io_poll is enabled |
| 273 | check_sysblock_value ${drive} "queue/wbt_lat_usec" 0 # Disabling wbt lat |
| 274 | show_device ${drive} |
| 275 | done |
| 276 | |
| 277 | check_poll_queue |
| 278 | compute_nb_threads ${drives} |
| 279 | check_scaling_governor |
| 280 | check_idle_governor |
| 281 | |
| 282 | info "##################################################" |
| 283 | echo |
| 284 | |
| 285 | cmdline="taskset -c ${taskset_cores} t/io_uring -b512 -d128 -c32 -s32 -p1 -F1 -B1 -n${nb_threads} ${latency_cmdline} ${drives}" |
| 286 | info "io_uring" "Running ${cmdline}" |
| 287 | ${cmdline} |