| 1 | #!/usr/bin/env bash
|
|---|
| 2 | # shellcheck disable=SC2013
|
|---|
| 3 | # shellcheck disable=SC2015
|
|---|
| 4 | # shellcheck disable=SC2034
|
|---|
| 5 | # shellcheck disable=SC2046
|
|---|
| 6 | # shellcheck disable=SC2059
|
|---|
| 7 | # shellcheck disable=SC2086 # we want word splitting
|
|---|
| 8 | # shellcheck disable=SC2154
|
|---|
| 9 | # shellcheck disable=SC2155
|
|---|
| 10 | # shellcheck disable=SC2162
|
|---|
| 11 | # shellcheck disable=SC2229
|
|---|
| 12 | #
|
|---|
| 13 | # This is an utility script to manage Intel GPU frequencies.
|
|---|
| 14 | # It can be used for debugging performance problems or trying to obtain a stable
|
|---|
| 15 | # frequency while benchmarking.
|
|---|
| 16 | #
|
|---|
| 17 | # Note the Intel i915 GPU driver allows to change the minimum, maximum and boost
|
|---|
| 18 | # frequencies in steps of 50 MHz via:
|
|---|
| 19 | #
|
|---|
| 20 | # /sys/class/drm/card<n>/<freq_info>
|
|---|
| 21 | #
|
|---|
| 22 | # Where <n> is the DRM card index and <freq_info> one of the following:
|
|---|
| 23 | #
|
|---|
| 24 | # - gt_max_freq_mhz (enforced maximum freq)
|
|---|
| 25 | # - gt_min_freq_mhz (enforced minimum freq)
|
|---|
| 26 | # - gt_boost_freq_mhz (enforced boost freq)
|
|---|
| 27 | #
|
|---|
| 28 | # The hardware capabilities can be accessed via:
|
|---|
| 29 | #
|
|---|
| 30 | # - gt_RP0_freq_mhz (supported maximum freq)
|
|---|
| 31 | # - gt_RPn_freq_mhz (supported minimum freq)
|
|---|
| 32 | # - gt_RP1_freq_mhz (most efficient freq)
|
|---|
| 33 | #
|
|---|
| 34 | # The current frequency can be read from:
|
|---|
| 35 | # - gt_act_freq_mhz (the actual GPU freq)
|
|---|
| 36 | # - gt_cur_freq_mhz (the last requested freq)
|
|---|
| 37 | #
|
|---|
| 38 | # Also note that in addition to GPU management, the script offers the
|
|---|
| 39 | # possibility to adjust CPU operating frequencies. However, this is currently
|
|---|
| 40 | # limited to just setting the maximum scaling frequency as percentage of the
|
|---|
| 41 | # maximum frequency allowed by the hardware.
|
|---|
| 42 | #
|
|---|
| 43 | # Copyright (C) 2022 Collabora Ltd.
|
|---|
| 44 | # Author: Cristian Ciocaltea <[email protected]>
|
|---|
| 45 | #
|
|---|
| 46 | # SPDX-License-Identifier: MIT
|
|---|
| 47 | #
|
|---|
| 48 |
|
|---|
| 49 | #
|
|---|
| 50 | # Constants
|
|---|
| 51 | #
|
|---|
| 52 |
|
|---|
| 53 | # GPU
|
|---|
| 54 | DRM_FREQ_SYSFS_PATTERN="/sys/class/drm/card%d/gt_%s_freq_mhz"
|
|---|
| 55 | ENF_FREQ_INFO="max min boost"
|
|---|
| 56 | CAP_FREQ_INFO="RP0 RPn RP1"
|
|---|
| 57 | ACT_FREQ_INFO="act cur"
|
|---|
| 58 | THROTT_DETECT_SLEEP_SEC=2
|
|---|
| 59 | THROTT_DETECT_PID_FILE_PATH=/tmp/thrott-detect.pid
|
|---|
| 60 |
|
|---|
| 61 | # CPU
|
|---|
| 62 | CPU_SYSFS_PREFIX=/sys/devices/system/cpu
|
|---|
| 63 | CPU_PSTATE_SYSFS_PATTERN="${CPU_SYSFS_PREFIX}/intel_pstate/%s"
|
|---|
| 64 | CPU_FREQ_SYSFS_PATTERN="${CPU_SYSFS_PREFIX}/cpu%s/cpufreq/%s_freq"
|
|---|
| 65 | CAP_CPU_FREQ_INFO="cpuinfo_max cpuinfo_min"
|
|---|
| 66 | ENF_CPU_FREQ_INFO="scaling_max scaling_min"
|
|---|
| 67 | ACT_CPU_FREQ_INFO="scaling_cur"
|
|---|
| 68 |
|
|---|
| 69 | #
|
|---|
| 70 | # Global variables.
|
|---|
| 71 | #
|
|---|
| 72 | unset INTEL_DRM_CARD_INDEX
|
|---|
| 73 | unset GET_ACT_FREQ GET_ENF_FREQ GET_CAP_FREQ
|
|---|
| 74 | unset SET_MIN_FREQ SET_MAX_FREQ
|
|---|
| 75 | unset MONITOR_FREQ
|
|---|
| 76 | unset CPU_SET_MAX_FREQ
|
|---|
| 77 | unset DETECT_THROTT
|
|---|
| 78 | unset DRY_RUN
|
|---|
| 79 |
|
|---|
| 80 | #
|
|---|
| 81 | # Simple printf based stderr logger.
|
|---|
| 82 | #
|
|---|
| 83 | log() {
|
|---|
| 84 | local msg_type=$1
|
|---|
| 85 |
|
|---|
| 86 | shift
|
|---|
| 87 | printf "%s: %s: " "${msg_type}" "${0##*/}" >&2
|
|---|
| 88 | printf "$@" >&2
|
|---|
| 89 | printf "\n" >&2
|
|---|
| 90 | }
|
|---|
| 91 |
|
|---|
| 92 | #
|
|---|
| 93 | # Helper to print sysfs path for the given card index and freq info.
|
|---|
| 94 | #
|
|---|
| 95 | # arg1: Frequency info sysfs name, one of *_FREQ_INFO constants above
|
|---|
| 96 | # arg2: Video card index, defaults to INTEL_DRM_CARD_INDEX
|
|---|
| 97 | #
|
|---|
| 98 | print_freq_sysfs_path() {
|
|---|
| 99 | printf ${DRM_FREQ_SYSFS_PATTERN} "${2:-${INTEL_DRM_CARD_INDEX}}" "$1"
|
|---|
| 100 | }
|
|---|
| 101 |
|
|---|
| 102 | #
|
|---|
| 103 | # Helper to set INTEL_DRM_CARD_INDEX for the first identified Intel video card.
|
|---|
| 104 | #
|
|---|
| 105 | identify_intel_gpu() {
|
|---|
| 106 | local i=0 vendor path
|
|---|
| 107 |
|
|---|
| 108 | while [ ${i} -lt 16 ]; do
|
|---|
| 109 | [ -c "/dev/dri/card$i" ] || {
|
|---|
| 110 | i=$((i + 1))
|
|---|
| 111 | continue
|
|---|
| 112 | }
|
|---|
| 113 |
|
|---|
| 114 | path=$(print_freq_sysfs_path "" ${i})
|
|---|
| 115 | path=${path%/*}/device/vendor
|
|---|
| 116 |
|
|---|
| 117 | [ -r "${path}" ] && read vendor < "${path}" && \
|
|---|
| 118 | [ "${vendor}" = "0x8086" ] && INTEL_DRM_CARD_INDEX=$i && return 0
|
|---|
| 119 |
|
|---|
| 120 | i=$((i + 1))
|
|---|
| 121 | done
|
|---|
| 122 |
|
|---|
| 123 | return 1
|
|---|
| 124 | }
|
|---|
| 125 |
|
|---|
| 126 | #
|
|---|
| 127 | # Read the specified freq info from sysfs.
|
|---|
| 128 | #
|
|---|
| 129 | # arg1: Flag (y/n) to also enable printing the freq info.
|
|---|
| 130 | # arg2...: Frequency info sysfs name(s), see *_FREQ_INFO constants above
|
|---|
| 131 | # return: Global variable(s) FREQ_${arg} containing the requested information
|
|---|
| 132 | #
|
|---|
| 133 | read_freq_info() {
|
|---|
| 134 | local var val info path print=0 ret=0
|
|---|
| 135 |
|
|---|
| 136 | [ "$1" = "y" ] && print=1
|
|---|
| 137 | shift
|
|---|
| 138 |
|
|---|
| 139 | while [ $# -gt 0 ]; do
|
|---|
| 140 | info=$1
|
|---|
| 141 | shift
|
|---|
| 142 | var=FREQ_${info}
|
|---|
| 143 | path=$(print_freq_sysfs_path "${info}")
|
|---|
| 144 |
|
|---|
| 145 | [ -r ${path} ] && read ${var} < ${path} || {
|
|---|
| 146 | log ERROR "Failed to read freq info from: %s" "${path}"
|
|---|
| 147 | ret=1
|
|---|
| 148 | continue
|
|---|
| 149 | }
|
|---|
| 150 |
|
|---|
| 151 | [ -n "${var}" ] || {
|
|---|
| 152 | log ERROR "Got empty freq info from: %s" "${path}"
|
|---|
| 153 | ret=1
|
|---|
| 154 | continue
|
|---|
| 155 | }
|
|---|
| 156 |
|
|---|
| 157 | [ ${print} -eq 1 ] && {
|
|---|
| 158 | eval val=\$${var}
|
|---|
| 159 | printf "%6s: %4s MHz\n" "${info}" "${val}"
|
|---|
| 160 | }
|
|---|
| 161 | done
|
|---|
| 162 |
|
|---|
| 163 | return ${ret}
|
|---|
| 164 | }
|
|---|
| 165 |
|
|---|
| 166 | #
|
|---|
| 167 | # Display requested info.
|
|---|
| 168 | #
|
|---|
| 169 | print_freq_info() {
|
|---|
| 170 | local req_freq
|
|---|
| 171 |
|
|---|
| 172 | [ -n "${GET_CAP_FREQ}" ] && {
|
|---|
| 173 | printf "* Hardware capabilities\n"
|
|---|
| 174 | read_freq_info y ${CAP_FREQ_INFO}
|
|---|
| 175 | printf "\n"
|
|---|
| 176 | }
|
|---|
| 177 |
|
|---|
| 178 | [ -n "${GET_ENF_FREQ}" ] && {
|
|---|
| 179 | printf "* Enforcements\n"
|
|---|
| 180 | read_freq_info y ${ENF_FREQ_INFO}
|
|---|
| 181 | printf "\n"
|
|---|
| 182 | }
|
|---|
| 183 |
|
|---|
| 184 | [ -n "${GET_ACT_FREQ}" ] && {
|
|---|
| 185 | printf "* Actual\n"
|
|---|
| 186 | read_freq_info y ${ACT_FREQ_INFO}
|
|---|
| 187 | printf "\n"
|
|---|
| 188 | }
|
|---|
| 189 | }
|
|---|
| 190 |
|
|---|
| 191 | #
|
|---|
| 192 | # Helper to print frequency value as requested by user via '-s, --set' option.
|
|---|
| 193 | # arg1: user requested freq value
|
|---|
| 194 | #
|
|---|
| 195 | compute_freq_set() {
|
|---|
| 196 | local val
|
|---|
| 197 |
|
|---|
| 198 | case "$1" in
|
|---|
| 199 | +)
|
|---|
| 200 | val=${FREQ_RP0}
|
|---|
| 201 | ;;
|
|---|
| 202 | -)
|
|---|
| 203 | val=${FREQ_RPn}
|
|---|
| 204 | ;;
|
|---|
| 205 | *%)
|
|---|
| 206 | val=$((${1%?} * FREQ_RP0 / 100))
|
|---|
| 207 | # Adjust freq to comply with 50 MHz increments
|
|---|
| 208 | val=$((val / 50 * 50))
|
|---|
| 209 | ;;
|
|---|
| 210 | *[!0-9]*)
|
|---|
| 211 | log ERROR "Cannot set freq to invalid value: %s" "$1"
|
|---|
| 212 | return 1
|
|---|
| 213 | ;;
|
|---|
| 214 | "")
|
|---|
| 215 | log ERROR "Cannot set freq to unspecified value"
|
|---|
| 216 | return 1
|
|---|
| 217 | ;;
|
|---|
| 218 | *)
|
|---|
| 219 | # Adjust freq to comply with 50 MHz increments
|
|---|
| 220 | val=$(($1 / 50 * 50))
|
|---|
| 221 | ;;
|
|---|
| 222 | esac
|
|---|
| 223 |
|
|---|
| 224 | printf "%s" "${val}"
|
|---|
| 225 | }
|
|---|
| 226 |
|
|---|
| 227 | #
|
|---|
| 228 | # Helper for set_freq().
|
|---|
| 229 | #
|
|---|
| 230 | set_freq_max() {
|
|---|
| 231 | log INFO "Setting GPU max freq to %s MHz" "${SET_MAX_FREQ}"
|
|---|
| 232 |
|
|---|
| 233 | read_freq_info n min || return $?
|
|---|
| 234 |
|
|---|
| 235 | [ ${SET_MAX_FREQ} -gt ${FREQ_RP0} ] && {
|
|---|
| 236 | log ERROR "Cannot set GPU max freq (%s) to be greater than hw max freq (%s)" \
|
|---|
| 237 | "${SET_MAX_FREQ}" "${FREQ_RP0}"
|
|---|
| 238 | return 1
|
|---|
| 239 | }
|
|---|
| 240 |
|
|---|
| 241 | [ ${SET_MAX_FREQ} -lt ${FREQ_RPn} ] && {
|
|---|
| 242 | log ERROR "Cannot set GPU max freq (%s) to be less than hw min freq (%s)" \
|
|---|
| 243 | "${SET_MIN_FREQ}" "${FREQ_RPn}"
|
|---|
| 244 | return 1
|
|---|
| 245 | }
|
|---|
| 246 |
|
|---|
| 247 | [ ${SET_MAX_FREQ} -lt ${FREQ_min} ] && {
|
|---|
| 248 | log ERROR "Cannot set GPU max freq (%s) to be less than min freq (%s)" \
|
|---|
| 249 | "${SET_MAX_FREQ}" "${FREQ_min}"
|
|---|
| 250 | return 1
|
|---|
| 251 | }
|
|---|
| 252 |
|
|---|
| 253 | [ -z "${DRY_RUN}" ] || return 0
|
|---|
| 254 |
|
|---|
| 255 | if ! printf "%s" ${SET_MAX_FREQ} | tee $(print_freq_sysfs_path max) \
|
|---|
| 256 | $(print_freq_sysfs_path boost) > /dev/null;
|
|---|
| 257 | then
|
|---|
| 258 | log ERROR "Failed to set GPU max frequency"
|
|---|
| 259 | return 1
|
|---|
| 260 | fi
|
|---|
| 261 | }
|
|---|
| 262 |
|
|---|
| 263 | #
|
|---|
| 264 | # Helper for set_freq().
|
|---|
| 265 | #
|
|---|
| 266 | set_freq_min() {
|
|---|
| 267 | log INFO "Setting GPU min freq to %s MHz" "${SET_MIN_FREQ}"
|
|---|
| 268 |
|
|---|
| 269 | read_freq_info n max || return $?
|
|---|
| 270 |
|
|---|
| 271 | [ ${SET_MIN_FREQ} -gt ${FREQ_max} ] && {
|
|---|
| 272 | log ERROR "Cannot set GPU min freq (%s) to be greater than max freq (%s)" \
|
|---|
| 273 | "${SET_MIN_FREQ}" "${FREQ_max}"
|
|---|
| 274 | return 1
|
|---|
| 275 | }
|
|---|
| 276 |
|
|---|
| 277 | [ ${SET_MIN_FREQ} -lt ${FREQ_RPn} ] && {
|
|---|
| 278 | log ERROR "Cannot set GPU min freq (%s) to be less than hw min freq (%s)" \
|
|---|
| 279 | "${SET_MIN_FREQ}" "${FREQ_RPn}"
|
|---|
| 280 | return 1
|
|---|
| 281 | }
|
|---|
| 282 |
|
|---|
| 283 | [ -z "${DRY_RUN}" ] || return 0
|
|---|
| 284 |
|
|---|
| 285 | if ! printf "%s" ${SET_MIN_FREQ} > $(print_freq_sysfs_path min);
|
|---|
| 286 | then
|
|---|
| 287 | log ERROR "Failed to set GPU min frequency"
|
|---|
| 288 | return 1
|
|---|
| 289 | fi
|
|---|
| 290 | }
|
|---|
| 291 |
|
|---|
| 292 | #
|
|---|
| 293 | # Set min or max or both GPU frequencies to the user indicated values.
|
|---|
| 294 | #
|
|---|
| 295 | set_freq() {
|
|---|
| 296 | # Get hw max & min frequencies
|
|---|
| 297 | read_freq_info n RP0 RPn || return $?
|
|---|
| 298 |
|
|---|
| 299 | [ -z "${SET_MAX_FREQ}" ] || {
|
|---|
| 300 | SET_MAX_FREQ=$(compute_freq_set "${SET_MAX_FREQ}")
|
|---|
| 301 | [ -z "${SET_MAX_FREQ}" ] && return 1
|
|---|
| 302 | }
|
|---|
| 303 |
|
|---|
| 304 | [ -z "${SET_MIN_FREQ}" ] || {
|
|---|
| 305 | SET_MIN_FREQ=$(compute_freq_set "${SET_MIN_FREQ}")
|
|---|
| 306 | [ -z "${SET_MIN_FREQ}" ] && return 1
|
|---|
| 307 | }
|
|---|
| 308 |
|
|---|
| 309 | #
|
|---|
| 310 | # Ensure correct operation order, to avoid setting min freq
|
|---|
| 311 | # to a value which is larger than max freq.
|
|---|
| 312 | #
|
|---|
| 313 | # E.g.:
|
|---|
| 314 | # crt_min=crt_max=600; new_min=new_max=700
|
|---|
| 315 | # > operation order: max=700; min=700
|
|---|
| 316 | #
|
|---|
| 317 | # crt_min=crt_max=600; new_min=new_max=500
|
|---|
| 318 | # > operation order: min=500; max=500
|
|---|
| 319 | #
|
|---|
| 320 | if [ -n "${SET_MAX_FREQ}" ] && [ -n "${SET_MIN_FREQ}" ]; then
|
|---|
| 321 | [ ${SET_MAX_FREQ} -lt ${SET_MIN_FREQ} ] && {
|
|---|
| 322 | log ERROR "Cannot set GPU max freq to be less than min freq"
|
|---|
| 323 | return 1
|
|---|
| 324 | }
|
|---|
| 325 |
|
|---|
| 326 | read_freq_info n min || return $?
|
|---|
| 327 |
|
|---|
| 328 | if [ ${SET_MAX_FREQ} -lt ${FREQ_min} ]; then
|
|---|
| 329 | set_freq_min || return $?
|
|---|
| 330 | set_freq_max
|
|---|
| 331 | else
|
|---|
| 332 | set_freq_max || return $?
|
|---|
| 333 | set_freq_min
|
|---|
| 334 | fi
|
|---|
| 335 | elif [ -n "${SET_MAX_FREQ}" ]; then
|
|---|
| 336 | set_freq_max
|
|---|
| 337 | elif [ -n "${SET_MIN_FREQ}" ]; then
|
|---|
| 338 | set_freq_min
|
|---|
| 339 | else
|
|---|
| 340 | log "Unexpected call to set_freq()"
|
|---|
| 341 | return 1
|
|---|
| 342 | fi
|
|---|
| 343 | }
|
|---|
| 344 |
|
|---|
| 345 | #
|
|---|
| 346 | # Helper for detect_throttling().
|
|---|
| 347 | #
|
|---|
| 348 | get_thrott_detect_pid() {
|
|---|
| 349 | [ -e ${THROTT_DETECT_PID_FILE_PATH} ] || return 0
|
|---|
| 350 |
|
|---|
| 351 | local pid
|
|---|
| 352 | read pid < ${THROTT_DETECT_PID_FILE_PATH} || {
|
|---|
| 353 | log ERROR "Failed to read pid from: %s" "${THROTT_DETECT_PID_FILE_PATH}"
|
|---|
| 354 | return 1
|
|---|
| 355 | }
|
|---|
| 356 |
|
|---|
| 357 | local proc_path=/proc/${pid:-invalid}/cmdline
|
|---|
| 358 | [ -r ${proc_path} ] && grep -qs "${0##*/}" ${proc_path} && {
|
|---|
| 359 | printf "%s" "${pid}"
|
|---|
| 360 | return 0
|
|---|
| 361 | }
|
|---|
| 362 |
|
|---|
| 363 | # Remove orphaned PID file
|
|---|
| 364 | rm -rf ${THROTT_DETECT_PID_FILE_PATH}
|
|---|
| 365 | return 1
|
|---|
| 366 | }
|
|---|
| 367 |
|
|---|
| 368 | #
|
|---|
| 369 | # Control detection and reporting of GPU throttling events.
|
|---|
| 370 | # arg1: start - run throttle detector in background
|
|---|
| 371 | # stop - stop throttle detector process, if any
|
|---|
| 372 | # status - verify if throttle detector is running
|
|---|
| 373 | #
|
|---|
| 374 | detect_throttling() {
|
|---|
| 375 | local pid
|
|---|
| 376 | pid=$(get_thrott_detect_pid)
|
|---|
| 377 |
|
|---|
| 378 | case "$1" in
|
|---|
| 379 | status)
|
|---|
| 380 | printf "Throttling detector is "
|
|---|
| 381 | [ -z "${pid}" ] && printf "not running\n" && return 0
|
|---|
| 382 | printf "running (pid=%s)\n" ${pid}
|
|---|
| 383 | ;;
|
|---|
| 384 |
|
|---|
| 385 | stop)
|
|---|
| 386 | [ -z "${pid}" ] && return 0
|
|---|
| 387 |
|
|---|
| 388 | log INFO "Stopping throttling detector (pid=%s)" "${pid}"
|
|---|
| 389 | kill ${pid}; sleep 1; kill -0 ${pid} 2>/dev/null && kill -9 ${pid}
|
|---|
| 390 | rm -rf ${THROTT_DETECT_PID_FILE_PATH}
|
|---|
| 391 | ;;
|
|---|
| 392 |
|
|---|
| 393 | start)
|
|---|
| 394 | [ -n "${pid}" ] && {
|
|---|
| 395 | log WARN "Throttling detector is already running (pid=%s)" ${pid}
|
|---|
| 396 | return 0
|
|---|
| 397 | }
|
|---|
| 398 |
|
|---|
| 399 | (
|
|---|
| 400 | read_freq_info n RPn || exit $?
|
|---|
| 401 |
|
|---|
| 402 | while true; do
|
|---|
| 403 | sleep ${THROTT_DETECT_SLEEP_SEC}
|
|---|
| 404 | read_freq_info n act min cur || exit $?
|
|---|
| 405 |
|
|---|
| 406 | #
|
|---|
| 407 | # The throttling seems to occur when act freq goes below min.
|
|---|
| 408 | # However, it's necessary to exclude the idle states, where
|
|---|
| 409 | # act freq normally reaches RPn and cur goes below min.
|
|---|
| 410 | #
|
|---|
| 411 | [ ${FREQ_act} -lt ${FREQ_min} ] && \
|
|---|
| 412 | [ ${FREQ_act} -gt ${FREQ_RPn} ] && \
|
|---|
| 413 | [ ${FREQ_cur} -ge ${FREQ_min} ] && \
|
|---|
| 414 | printf "GPU throttling detected: act=%s min=%s cur=%s RPn=%s\n" \
|
|---|
| 415 | ${FREQ_act} ${FREQ_min} ${FREQ_cur} ${FREQ_RPn}
|
|---|
| 416 | done
|
|---|
| 417 | ) &
|
|---|
| 418 |
|
|---|
| 419 | pid=$!
|
|---|
| 420 | log INFO "Started GPU throttling detector (pid=%s)" ${pid}
|
|---|
| 421 |
|
|---|
| 422 | printf "%s\n" ${pid} > ${THROTT_DETECT_PID_FILE_PATH} || \
|
|---|
| 423 | log WARN "Failed to write throttle detector PID file"
|
|---|
| 424 | ;;
|
|---|
| 425 | esac
|
|---|
| 426 | }
|
|---|
| 427 |
|
|---|
| 428 | #
|
|---|
| 429 | # Retrieve the list of online CPUs.
|
|---|
| 430 | #
|
|---|
| 431 | get_online_cpus() {
|
|---|
| 432 | local path cpu_index
|
|---|
| 433 |
|
|---|
| 434 | printf "0"
|
|---|
| 435 | for path in $(grep 1 ${CPU_SYSFS_PREFIX}/cpu*/online); do
|
|---|
| 436 | cpu_index=${path##*/cpu}
|
|---|
| 437 | printf " %s" ${cpu_index%%/*}
|
|---|
| 438 | done
|
|---|
| 439 | }
|
|---|
| 440 |
|
|---|
| 441 | #
|
|---|
| 442 | # Helper to print sysfs path for the given CPU index and freq info.
|
|---|
| 443 | #
|
|---|
| 444 | # arg1: Frequency info sysfs name, one of *_CPU_FREQ_INFO constants above
|
|---|
| 445 | # arg2: CPU index
|
|---|
| 446 | #
|
|---|
| 447 | print_cpu_freq_sysfs_path() {
|
|---|
| 448 | printf ${CPU_FREQ_SYSFS_PATTERN} "$2" "$1"
|
|---|
| 449 | }
|
|---|
| 450 |
|
|---|
| 451 | #
|
|---|
| 452 | # Read the specified CPU freq info from sysfs.
|
|---|
| 453 | #
|
|---|
| 454 | # arg1: CPU index
|
|---|
| 455 | # arg2: Flag (y/n) to also enable printing the freq info.
|
|---|
| 456 | # arg3...: Frequency info sysfs name(s), see *_CPU_FREQ_INFO constants above
|
|---|
| 457 | # return: Global variable(s) CPU_FREQ_${arg} containing the requested information
|
|---|
| 458 | #
|
|---|
| 459 | read_cpu_freq_info() {
|
|---|
| 460 | local var val info path cpu_index print=0 ret=0
|
|---|
| 461 |
|
|---|
| 462 | cpu_index=$1
|
|---|
| 463 | [ "$2" = "y" ] && print=1
|
|---|
| 464 | shift 2
|
|---|
| 465 |
|
|---|
| 466 | while [ $# -gt 0 ]; do
|
|---|
| 467 | info=$1
|
|---|
| 468 | shift
|
|---|
| 469 | var=CPU_FREQ_${info}
|
|---|
| 470 | path=$(print_cpu_freq_sysfs_path "${info}" ${cpu_index})
|
|---|
| 471 |
|
|---|
| 472 | [ -r ${path} ] && read ${var} < ${path} || {
|
|---|
| 473 | log ERROR "Failed to read CPU freq info from: %s" "${path}"
|
|---|
| 474 | ret=1
|
|---|
| 475 | continue
|
|---|
| 476 | }
|
|---|
| 477 |
|
|---|
| 478 | [ -n "${var}" ] || {
|
|---|
| 479 | log ERROR "Got empty CPU freq info from: %s" "${path}"
|
|---|
| 480 | ret=1
|
|---|
| 481 | continue
|
|---|
| 482 | }
|
|---|
| 483 |
|
|---|
| 484 | [ ${print} -eq 1 ] && {
|
|---|
| 485 | eval val=\$${var}
|
|---|
| 486 | printf "%6s: %4s Hz\n" "${info}" "${val}"
|
|---|
| 487 | }
|
|---|
| 488 | done
|
|---|
| 489 |
|
|---|
| 490 | return ${ret}
|
|---|
| 491 | }
|
|---|
| 492 |
|
|---|
| 493 | #
|
|---|
| 494 | # Helper to print freq. value as requested by user via '--cpu-set-max' option.
|
|---|
| 495 | # arg1: user requested freq value
|
|---|
| 496 | #
|
|---|
| 497 | compute_cpu_freq_set() {
|
|---|
| 498 | local val
|
|---|
| 499 |
|
|---|
| 500 | case "$1" in
|
|---|
| 501 | +)
|
|---|
| 502 | val=${CPU_FREQ_cpuinfo_max}
|
|---|
| 503 | ;;
|
|---|
| 504 | -)
|
|---|
| 505 | val=${CPU_FREQ_cpuinfo_min}
|
|---|
| 506 | ;;
|
|---|
| 507 | *%)
|
|---|
| 508 | val=$((${1%?} * CPU_FREQ_cpuinfo_max / 100))
|
|---|
| 509 | ;;
|
|---|
| 510 | *[!0-9]*)
|
|---|
| 511 | log ERROR "Cannot set CPU freq to invalid value: %s" "$1"
|
|---|
| 512 | return 1
|
|---|
| 513 | ;;
|
|---|
| 514 | "")
|
|---|
| 515 | log ERROR "Cannot set CPU freq to unspecified value"
|
|---|
| 516 | return 1
|
|---|
| 517 | ;;
|
|---|
| 518 | *)
|
|---|
| 519 | log ERROR "Cannot set CPU freq to custom value; use +, -, or % instead"
|
|---|
| 520 | return 1
|
|---|
| 521 | ;;
|
|---|
| 522 | esac
|
|---|
| 523 |
|
|---|
| 524 | printf "%s" "${val}"
|
|---|
| 525 | }
|
|---|
| 526 |
|
|---|
| 527 | #
|
|---|
| 528 | # Adjust CPU max scaling frequency.
|
|---|
| 529 | #
|
|---|
| 530 | set_cpu_freq_max() {
|
|---|
| 531 | local target_freq res=0
|
|---|
| 532 | case "${CPU_SET_MAX_FREQ}" in
|
|---|
| 533 | +)
|
|---|
| 534 | target_freq=100
|
|---|
| 535 | ;;
|
|---|
| 536 | -)
|
|---|
| 537 | target_freq=1
|
|---|
| 538 | ;;
|
|---|
| 539 | *%)
|
|---|
| 540 | target_freq=${CPU_SET_MAX_FREQ%?}
|
|---|
| 541 | ;;
|
|---|
| 542 | *)
|
|---|
| 543 | log ERROR "Invalid CPU freq"
|
|---|
| 544 | return 1
|
|---|
| 545 | ;;
|
|---|
| 546 | esac
|
|---|
| 547 |
|
|---|
| 548 | local pstate_info=$(printf "${CPU_PSTATE_SYSFS_PATTERN}" max_perf_pct)
|
|---|
| 549 | [ -e "${pstate_info}" ] && {
|
|---|
| 550 | log INFO "Setting intel_pstate max perf to %s" "${target_freq}%"
|
|---|
| 551 | if ! printf "%s" "${target_freq}" > "${pstate_info}";
|
|---|
| 552 | then
|
|---|
| 553 | log ERROR "Failed to set intel_pstate max perf"
|
|---|
| 554 | res=1
|
|---|
| 555 | fi
|
|---|
| 556 | }
|
|---|
| 557 |
|
|---|
| 558 | local cpu_index
|
|---|
| 559 | for cpu_index in $(get_online_cpus); do
|
|---|
| 560 | read_cpu_freq_info ${cpu_index} n ${CAP_CPU_FREQ_INFO} || { res=$?; continue; }
|
|---|
| 561 |
|
|---|
| 562 | target_freq=$(compute_cpu_freq_set "${CPU_SET_MAX_FREQ}")
|
|---|
| 563 | [ -z "${target_freq}" ] && { res=$?; continue; }
|
|---|
| 564 |
|
|---|
| 565 | log INFO "Setting CPU%s max scaling freq to %s Hz" ${cpu_index} "${target_freq}"
|
|---|
| 566 | [ -n "${DRY_RUN}" ] && continue
|
|---|
| 567 |
|
|---|
| 568 | if ! printf "%s" ${target_freq} > $(print_cpu_freq_sysfs_path scaling_max ${cpu_index});
|
|---|
| 569 | then
|
|---|
| 570 | res=1
|
|---|
| 571 | log ERROR "Failed to set CPU%s max scaling frequency" ${cpu_index}
|
|---|
| 572 | fi
|
|---|
| 573 | done
|
|---|
| 574 |
|
|---|
| 575 | return ${res}
|
|---|
| 576 | }
|
|---|
| 577 |
|
|---|
| 578 | #
|
|---|
| 579 | # Show help message.
|
|---|
| 580 | #
|
|---|
| 581 | print_usage() {
|
|---|
| 582 | cat <<EOF
|
|---|
| 583 | Usage: ${0##*/} [OPTION]...
|
|---|
| 584 |
|
|---|
| 585 | A script to manage Intel GPU frequencies. Can be used for debugging performance
|
|---|
| 586 | problems or trying to obtain a stable frequency while benchmarking.
|
|---|
| 587 |
|
|---|
| 588 | Note Intel GPUs only accept specific frequencies, usually multiples of 50 MHz.
|
|---|
| 589 |
|
|---|
| 590 | Options:
|
|---|
| 591 | -g, --get [act|enf|cap|all]
|
|---|
| 592 | Get frequency information: active (default), enforced,
|
|---|
| 593 | hardware capabilities or all of them.
|
|---|
| 594 |
|
|---|
| 595 | -s, --set [{min|max}=]{FREQUENCY[%]|+|-}
|
|---|
| 596 | Set min or max frequency to the given value (MHz).
|
|---|
| 597 | Append '%' to interpret FREQUENCY as % of hw max.
|
|---|
| 598 | Use '+' or '-' to set frequency to hardware max or min.
|
|---|
| 599 | Omit min/max prefix to set both frequencies.
|
|---|
| 600 |
|
|---|
| 601 | -r, --reset Reset frequencies to hardware defaults.
|
|---|
| 602 |
|
|---|
| 603 | -m, --monitor [act|enf|cap|all]
|
|---|
| 604 | Monitor the indicated frequencies via 'watch' utility.
|
|---|
| 605 | See '-g, --get' option for more details.
|
|---|
| 606 |
|
|---|
| 607 | -d|--detect-thrott [start|stop|status]
|
|---|
| 608 | Start (default operation) the throttling detector
|
|---|
| 609 | as a background process. Use 'stop' or 'status' to
|
|---|
| 610 | terminate the detector process or verify its status.
|
|---|
| 611 |
|
|---|
| 612 | --cpu-set-max [FREQUENCY%|+|-}
|
|---|
| 613 | Set CPU max scaling frequency as % of hw max.
|
|---|
| 614 | Use '+' or '-' to set frequency to hardware max or min.
|
|---|
| 615 |
|
|---|
| 616 | -r, --reset Reset frequencies to hardware defaults.
|
|---|
| 617 |
|
|---|
| 618 | --dry-run See what the script will do without applying any
|
|---|
| 619 | frequency changes.
|
|---|
| 620 |
|
|---|
| 621 | -h, --help Display this help text and exit.
|
|---|
| 622 | EOF
|
|---|
| 623 | }
|
|---|
| 624 |
|
|---|
| 625 | #
|
|---|
| 626 | # Parse user input for '-g, --get' option.
|
|---|
| 627 | # Returns 0 if a value has been provided, otherwise 1.
|
|---|
| 628 | #
|
|---|
| 629 | parse_option_get() {
|
|---|
| 630 | local ret=0
|
|---|
| 631 |
|
|---|
| 632 | case "$1" in
|
|---|
| 633 | act) GET_ACT_FREQ=1;;
|
|---|
| 634 | enf) GET_ENF_FREQ=1;;
|
|---|
| 635 | cap) GET_CAP_FREQ=1;;
|
|---|
| 636 | all) GET_ACT_FREQ=1; GET_ENF_FREQ=1; GET_CAP_FREQ=1;;
|
|---|
| 637 | -*|"")
|
|---|
| 638 | # No value provided, using default.
|
|---|
| 639 | GET_ACT_FREQ=1
|
|---|
| 640 | ret=1
|
|---|
| 641 | ;;
|
|---|
| 642 | *)
|
|---|
| 643 | print_usage
|
|---|
| 644 | exit 1
|
|---|
| 645 | ;;
|
|---|
| 646 | esac
|
|---|
| 647 |
|
|---|
| 648 | return ${ret}
|
|---|
| 649 | }
|
|---|
| 650 |
|
|---|
| 651 | #
|
|---|
| 652 | # Validate user input for '-s, --set' option.
|
|---|
| 653 | # arg1: input value to be validated
|
|---|
| 654 | # arg2: optional flag indicating input is restricted to %
|
|---|
| 655 | #
|
|---|
| 656 | validate_option_set() {
|
|---|
| 657 | case "$1" in
|
|---|
| 658 | +|-|[0-9]%|[0-9][0-9]%)
|
|---|
| 659 | return 0
|
|---|
| 660 | ;;
|
|---|
| 661 | *[!0-9]*|"")
|
|---|
| 662 | print_usage
|
|---|
| 663 | exit 1
|
|---|
| 664 | ;;
|
|---|
| 665 | esac
|
|---|
| 666 |
|
|---|
| 667 | [ -z "$2" ] || { print_usage; exit 1; }
|
|---|
| 668 | }
|
|---|
| 669 |
|
|---|
| 670 | #
|
|---|
| 671 | # Parse script arguments.
|
|---|
| 672 | #
|
|---|
| 673 | [ $# -eq 0 ] && { print_usage; exit 1; }
|
|---|
| 674 |
|
|---|
| 675 | while [ $# -gt 0 ]; do
|
|---|
| 676 | case "$1" in
|
|---|
| 677 | -g|--get)
|
|---|
| 678 | parse_option_get "$2" && shift
|
|---|
| 679 | ;;
|
|---|
| 680 |
|
|---|
| 681 | -s|--set)
|
|---|
| 682 | shift
|
|---|
| 683 | case "$1" in
|
|---|
| 684 | min=*)
|
|---|
| 685 | SET_MIN_FREQ=${1#min=}
|
|---|
| 686 | validate_option_set "${SET_MIN_FREQ}"
|
|---|
| 687 | ;;
|
|---|
| 688 | max=*)
|
|---|
| 689 | SET_MAX_FREQ=${1#max=}
|
|---|
| 690 | validate_option_set "${SET_MAX_FREQ}"
|
|---|
| 691 | ;;
|
|---|
| 692 | *)
|
|---|
| 693 | SET_MIN_FREQ=$1
|
|---|
| 694 | validate_option_set "${SET_MIN_FREQ}"
|
|---|
| 695 | SET_MAX_FREQ=${SET_MIN_FREQ}
|
|---|
| 696 | ;;
|
|---|
| 697 | esac
|
|---|
| 698 | ;;
|
|---|
| 699 |
|
|---|
| 700 | -r|--reset)
|
|---|
| 701 | RESET_FREQ=1
|
|---|
| 702 | SET_MIN_FREQ="-"
|
|---|
| 703 | SET_MAX_FREQ="+"
|
|---|
| 704 | ;;
|
|---|
| 705 |
|
|---|
| 706 | -m|--monitor)
|
|---|
| 707 | MONITOR_FREQ=act
|
|---|
| 708 | parse_option_get "$2" && MONITOR_FREQ=$2 && shift
|
|---|
| 709 | ;;
|
|---|
| 710 |
|
|---|
| 711 | -d|--detect-thrott)
|
|---|
| 712 | DETECT_THROTT=start
|
|---|
| 713 | case "$2" in
|
|---|
| 714 | start|stop|status)
|
|---|
| 715 | DETECT_THROTT=$2
|
|---|
| 716 | shift
|
|---|
| 717 | ;;
|
|---|
| 718 | esac
|
|---|
| 719 | ;;
|
|---|
| 720 |
|
|---|
| 721 | --cpu-set-max)
|
|---|
| 722 | shift
|
|---|
| 723 | CPU_SET_MAX_FREQ=$1
|
|---|
| 724 | validate_option_set "${CPU_SET_MAX_FREQ}" restricted
|
|---|
| 725 | ;;
|
|---|
| 726 |
|
|---|
| 727 | --dry-run)
|
|---|
| 728 | DRY_RUN=1
|
|---|
| 729 | ;;
|
|---|
| 730 |
|
|---|
| 731 | -h|--help)
|
|---|
| 732 | print_usage
|
|---|
| 733 | exit 0
|
|---|
| 734 | ;;
|
|---|
| 735 |
|
|---|
| 736 | *)
|
|---|
| 737 | print_usage
|
|---|
| 738 | exit 1
|
|---|
| 739 | ;;
|
|---|
| 740 | esac
|
|---|
| 741 |
|
|---|
| 742 | shift
|
|---|
| 743 | done
|
|---|
| 744 |
|
|---|
| 745 | #
|
|---|
| 746 | # Main
|
|---|
| 747 | #
|
|---|
| 748 | RET=0
|
|---|
| 749 |
|
|---|
| 750 | identify_intel_gpu || {
|
|---|
| 751 | log INFO "No Intel GPU detected"
|
|---|
| 752 | exit 0
|
|---|
| 753 | }
|
|---|
| 754 |
|
|---|
| 755 | [ -n "${SET_MIN_FREQ}${SET_MAX_FREQ}" ] && { set_freq || RET=$?; }
|
|---|
| 756 | print_freq_info
|
|---|
| 757 |
|
|---|
| 758 | [ -n "${DETECT_THROTT}" ] && detect_throttling ${DETECT_THROTT}
|
|---|
| 759 |
|
|---|
| 760 | [ -n "${CPU_SET_MAX_FREQ}" ] && { set_cpu_freq_max || RET=$?; }
|
|---|
| 761 |
|
|---|
| 762 | [ -n "${MONITOR_FREQ}" ] && {
|
|---|
| 763 | log INFO "Entering frequency monitoring mode"
|
|---|
| 764 | sleep 2
|
|---|
| 765 | exec watch -d -n 1 "$0" -g "${MONITOR_FREQ}"
|
|---|
| 766 | }
|
|---|
| 767 |
|
|---|
| 768 | exit ${RET}
|
|---|
