Documentation/ABI/testing/sysfs-devices-system-cpu

   1 What:           /sys/devices/system/cpu/
   2 Date:           pre-git history
   3 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
   4 Description:
   5                 A collection of both global and individual CPU attributes
   6
   7                 Individual CPU attributes are contained in subdirectories
   8                 named by the kernel's logical CPU number, e.g.:
   9
  10                 /sys/devices/system/cpu/cpu#/
  11
  12 What:           /sys/devices/system/cpu/kernel_max
  13                 /sys/devices/system/cpu/offline
  14                 /sys/devices/system/cpu/online
  15                 /sys/devices/system/cpu/possible
  16                 /sys/devices/system/cpu/present
  17 Date:           December 2008
  18 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
  19 Description:    CPU topology files that describe kernel limits related to
  20                 hotplug. Briefly:
  21
  22                 kernel_max: the maximum cpu index allowed by the kernel
  23                 configuration.
  24
  25                 offline: cpus that are not online because they have been
  26                 HOTPLUGGED off or exceed the limit of cpus allowed by the
  27                 kernel configuration (kernel_max above).
  28
  29                 online: cpus that are online and being scheduled.
  30
  31                 possible: cpus that have been allocated resources and can be
  32                 brought online if they are present.
  33
  34                 present: cpus that have been identified as being present in
  35                 the system.
  36
  37                 See Documentation/cputopology.txt for more information.
  38
  39
  40 What:           /sys/devices/system/cpu/probe
  41                 /sys/devices/system/cpu/release
  42 Date:           November 2009
  43 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
  44 Description:    Dynamic addition and removal of CPU's.  This is not hotplug
  45                 removal, this is meant complete removal/addition of the CPU
  46                 from the system.
  47
  48                 probe: writes to this file will dynamically add a CPU to the
  49                 system.  Information written to the file to add CPU's is
  50                 architecture specific.
  51
  52                 release: writes to this file dynamically remove a CPU from
  53                 the system.  Information writtento the file to remove CPU's
  54                 is architecture specific.
  55
  56 What:           /sys/devices/system/cpu/cpu#/node
  57 Date:           October 2009
  58 Contact:        Linux memory management mailing list <linux-mm@kvack.org>
  59 Description:    Discover NUMA node a CPU belongs to
  60
  61                 When CONFIG_NUMA is enabled, a symbolic link that points
  62                 to the corresponding NUMA node directory.
  63
  64                 For example, the following symlink is created for cpu42
  65                 in NUMA node 2:
  66
  67                 /sys/devices/system/cpu/cpu42/node2 -> ../../node/node2
  68
  69
  70 What:           /sys/devices/system/cpu/cpu#/topology/core_id
  71                 /sys/devices/system/cpu/cpu#/topology/core_siblings
  72                 /sys/devices/system/cpu/cpu#/topology/core_siblings_list
  73                 /sys/devices/system/cpu/cpu#/topology/physical_package_id
  74                 /sys/devices/system/cpu/cpu#/topology/thread_siblings
  75                 /sys/devices/system/cpu/cpu#/topology/thread_siblings_list
  76 Date:           December 2008
  77 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
  78 Description:    CPU topology files that describe a logical CPU's relationship
  79                 to other cores and threads in the same physical package.
  80
  81                 One cpu# directory is created per logical CPU in the system,
  82                 e.g. /sys/devices/system/cpu/cpu42/.
  83
  84                 Briefly, the files above are:
  85
  86                 core_id: the CPU core ID of cpu#. Typically it is the
  87                 hardware platform's identifier (rather than the kernel's).
  88                 The actual value is architecture and platform dependent.
  89
  90                 core_siblings: internal kernel map of cpu#'s hardware threads
  91                 within the same physical_package_id.
  92
  93                 core_siblings_list: human-readable list of the logical CPU
  94                 numbers within the same physical_package_id as cpu#.
  95
  96                 physical_package_id: physical package id of cpu#. Typically
  97                 corresponds to a physical socket number, but the actual value
  98                 is architecture and platform dependent.
  99
 100                 thread_siblings: internel kernel map of cpu#'s hardware
 101                 threads within the same core as cpu#
 102
 103                 thread_siblings_list: human-readable list of cpu#'s hardware
 104                 threads within the same core as cpu#
 105
 106                 See Documentation/cputopology.txt for more information.
 107
 108
 109 What:           /sys/devices/system/cpu/cpuidle/current_driver
 110                 /sys/devices/system/cpu/cpuidle/current_governer_ro
 111                 /sys/devices/system/cpu/cpuidle/available_governors
 112                 /sys/devices/system/cpu/cpuidle/current_governor
 113 Date:           September 2007
 114 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 115 Description:    Discover cpuidle policy and mechanism
 116
 117                 Various CPUs today support multiple idle levels that are
 118                 differentiated by varying exit latencies and power
 119                 consumption during idle.
 120
 121                 Idle policy (governor) is differentiated from idle mechanism
 122                 (driver)
 123
 124                 current_driver: (RO) displays current idle mechanism
 125
 126                 current_governor_ro: (RO) displays current idle policy
 127
 128                 With the cpuidle_sysfs_switch boot option enabled (meant for
 129                 developer testing), the following three attributes are visible
 130                 instead:
 131
 132                 current_driver: same as described above
 133
 134                 available_governors: (RO) displays a space separated list of
 135                 available governors
 136
 137                 current_governor: (RW) displays current idle policy. Users can
 138                 switch the governor at runtime by writing to this file.
 139
 140                 See files in Documentation/cpuidle/ for more information.
 141
 142
 143 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/name
 144                 /sys/devices/system/cpu/cpuX/cpuidle/stateN/latency
 145                 /sys/devices/system/cpu/cpuX/cpuidle/stateN/power
 146                 /sys/devices/system/cpu/cpuX/cpuidle/stateN/time
 147                 /sys/devices/system/cpu/cpuX/cpuidle/stateN/usage
 148                 /sys/devices/system/cpu/cpuX/cpuidle/stateN/above
 149                 /sys/devices/system/cpu/cpuX/cpuidle/stateN/below
 150 Date:           September 2007
 151 KernelVersion:  v2.6.24
 152 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 153 Description:
 154                 The directory /sys/devices/system/cpu/cpuX/cpuidle contains per
 155                 logical CPU specific cpuidle information for each online cpu X.
 156                 The processor idle states which are available for use have the
 157                 following attributes:
 158
 159                 name: (RO) Name of the idle state (string).
 160
 161                 latency: (RO) The latency to exit out of this idle state (in
 162                 microseconds).
 163
 164                 power: (RO) The power consumed while in this idle state (in
 165                 milliwatts).
 166
 167                 time: (RO) The total time spent in this idle state (in microseconds).
 168
 169                 usage: (RO) Number of times this state was entered (a count).
 170
 171                 above: (RO) Number of times this state was entered, but the
 172                        observed CPU idle duration was too short for it (a count).
 173
 174                 below: (RO) Number of times this state was entered, but the
 175                        observed CPU idle duration was too long for it (a count).
 176
 177 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/desc
 178 Date:           February 2008
 179 KernelVersion:  v2.6.25
 180 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 181 Description:
 182                 (RO) A small description about the idle state (string).
 183
 184
 185 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/disable
 186 Date:           March 2012
 187 KernelVersion:  v3.10
 188 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 189 Description:
 190                 (RW) Option to disable this idle state (bool). The behavior and
 191                 the effect of the disable variable depends on the implementation
 192                 of a particular governor. In the ladder governor, for example,
 193                 it is not coherent, i.e. if one is disabling a light state, then
 194                 all deeper states are disabled as well, but the disable variable
 195                 does not reflect it. Likewise, if one enables a deep state but a
 196                 lighter state still is disabled, then this has no effect.
 197
 198
 199 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/residency
 200 Date:           March 2014
 201 KernelVersion:  v3.15
 202 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 203 Description:
 204                 (RO) Display the target residency i.e. the minimum amount of
 205                 time (in microseconds) this cpu should spend in this idle state
 206                 to make the transition worth the effort.
 207
 208 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/s2idle/
 209 Date:           March 2018
 210 KernelVersion:  v4.17
 211 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 212 Description:
 213                 Idle state usage statistics related to suspend-to-idle.
 214
 215                 This attribute group is only present for states that can be
 216                 used in suspend-to-idle with suspended timekeeping.
 217
 218 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/s2idle/time
 219 Date:           March 2018
 220 KernelVersion:  v4.17
 221 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 222 Description:
 223                 Total time spent by the CPU in suspend-to-idle (with scheduler
 224                 tick suspended) after requesting this state.
 225
 226 What:           /sys/devices/system/cpu/cpuX/cpuidle/stateN/s2idle/usage
 227 Date:           March 2018
 228 KernelVersion:  v4.17
 229 Contact:        Linux power management list <linux-pm@vger.kernel.org>
 230 Description:
 231                 Total number of times this state has been requested by the CPU
 232                 while entering suspend-to-idle.
 233
 234 What:           /sys/devices/system/cpu/cpu#/cpufreq/*
 235 Date:           pre-git history
 236 Contact:        linux-pm@vger.kernel.org
 237 Description:    Discover and change clock speed of CPUs
 238
 239                 Clock scaling allows you to change the clock speed of the
 240                 CPUs on the fly. This is a nice method to save battery
 241                 power, because the lower the clock speed, the less power
 242                 the CPU consumes.
 243
 244                 There are many knobs to tweak in this directory.
 245
 246                 See files in Documentation/cpu-freq/ for more information.
 247
 248
 249 What:           /sys/devices/system/cpu/cpu#/cpufreq/freqdomain_cpus
 250 Date:           June 2013
 251 Contact:        linux-pm@vger.kernel.org
 252 Description:    Discover CPUs in the same CPU frequency coordination domain
 253
 254                 freqdomain_cpus is the list of CPUs (online+offline) that share
 255                 the same clock/freq domain (possibly at the hardware level).
 256                 That information may be hidden from the cpufreq core and the
 257                 value of related_cpus may be different from freqdomain_cpus. This
 258                 attribute is useful for user space DVFS controllers to get better
 259                 power/performance results for platforms using acpi-cpufreq.
 260
 261                 This file is only present if the acpi-cpufreq driver is in use.
 262
 263
 264 What:           /sys/devices/system/cpu/cpu*/cache/index3/cache_disable_{0,1}
 265 Date:           August 2008
 266 KernelVersion:  2.6.27
 267 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 268 Description:    Disable L3 cache indices
 269
 270                 These files exist in every CPU's cache/index3 directory. Each
 271                 cache_disable_{0,1} file corresponds to one disable slot which
 272                 can be used to disable a cache index. Reading from these files
 273                 on a processor with this functionality will return the currently
 274                 disabled index for that node. There is one L3 structure per
 275                 node, or per internal node on MCM machines. Writing a valid
 276                 index to one of these files will cause the specificed cache
 277                 index to be disabled.
 278
 279                 All AMD processors with L3 caches provide this functionality.
 280                 For details, see BKDGs at
 281                 http://developer.amd.com/documentation/guides/Pages/default.aspx
 282
 283
 284 What:           /sys/devices/system/cpu/cpufreq/boost
 285 Date:           August 2012
 286 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 287 Description:    Processor frequency boosting control
 288
 289                 This switch controls the boost setting for the whole system.
 290                 Boosting allows the CPU and the firmware to run at a frequency
 291                 beyound it's nominal limit.
 292                 More details can be found in
 293                 Documentation/admin-guide/pm/cpufreq.rst
 294
 295
 296 What:           /sys/devices/system/cpu/cpu#/crash_notes
 297                 /sys/devices/system/cpu/cpu#/crash_notes_size
 298 Date:           April 2013
 299 Contact:        kexec@lists.infradead.org
 300 Description:    address and size of the percpu note.
 301
 302                 crash_notes: the physical address of the memory that holds the
 303                 note of cpu#.
 304
 305                 crash_notes_size: size of the note of cpu#.
 306
 307
 308 What:           /sys/devices/system/cpu/intel_pstate/max_perf_pct
 309                 /sys/devices/system/cpu/intel_pstate/min_perf_pct
 310                 /sys/devices/system/cpu/intel_pstate/no_turbo
 311 Date:           February 2013
 312 Contact:        linux-pm@vger.kernel.org
 313 Description:    Parameters for the Intel P-state driver
 314
 315                 Logic for selecting the current P-state in Intel
 316                 Sandybridge+ processors. The three knobs control
 317                 limits for the P-state that will be requested by the
 318                 driver.
 319
 320                 max_perf_pct: limits the maximum P state that will be requested by
 321                 the driver stated as a percentage of the available performance.
 322
 323                 min_perf_pct: limits the minimum P state that will be requested by
 324                 the driver stated as a percentage of the available performance.
 325
 326                 no_turbo: limits the driver to selecting P states below the turbo
 327                 frequency range.
 328
 329                 More details can be found in
 330                 Documentation/admin-guide/pm/intel_pstate.rst
 331
 332 What:           /sys/devices/system/cpu/cpu*/cache/index*/<set_of_attributes_mentioned_below>
 333 Date:           July 2014(documented, existed before August 2008)
 334 Contact:        Sudeep Holla <sudeep.holla@arm.com>
 335                 Linux kernel mailing list <linux-kernel@vger.kernel.org>
 336 Description:    Parameters for the CPU cache attributes
 337
 338                 allocation_policy:
 339                         - WriteAllocate: allocate a memory location to a cache line
 340                                          on a cache miss because of a write
 341                         - ReadAllocate: allocate a memory location to a cache line
 342                                         on a cache miss because of a read
 343                         - ReadWriteAllocate: both writeallocate and readallocate
 344
 345                 attributes: LEGACY used only on IA64 and is same as write_policy
 346
 347                 coherency_line_size: the minimum amount of data in bytes that gets
 348                                      transferred from memory to cache
 349
 350                 level: the cache hierarchy in the multi-level cache configuration
 351
 352                 number_of_sets: total number of sets in the cache, a set is a
 353                                 collection of cache lines with the same cache index
 354
 355                 physical_line_partition: number of physical cache line per cache tag
 356
 357                 shared_cpu_list: the list of logical cpus sharing the cache
 358
 359                 shared_cpu_map: logical cpu mask containing the list of cpus sharing
 360                                 the cache
 361
 362                 size: the total cache size in kB
 363
 364                 type:
 365                         - Instruction: cache that only holds instructions
 366                         - Data: cache that only caches data
 367                         - Unified: cache that holds both data and instructions
 368
 369                 ways_of_associativity: degree of freedom in placing a particular block
 370                                         of memory in the cache
 371
 372                 write_policy:
 373                         - WriteThrough: data is written to both the cache line
 374                                         and to the block in the lower-level memory
 375                         - WriteBack: data is written only to the cache line and
 376                                      the modified cache line is written to main
 377                                      memory only when it is replaced
 378
 379
 380 What:           /sys/devices/system/cpu/cpu*/cache/index*/id
 381 Date:           September 2016
 382 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 383 Description:    Cache id
 384
 385                 The id provides a unique number for a specific instance of
 386                 a cache of a particular type. E.g. there may be a level
 387                 3 unified cache on each socket in a server and we may
 388                 assign them ids 0, 1, 2, ...
 389
 390                 Note that id value can be non-contiguous. E.g. level 1
 391                 caches typically exist per core, but there may not be a
 392                 power of two cores on a socket, so these caches may be
 393                 numbered 0, 1, 2, 3, 4, 5, 8, 9, 10, ...
 394
 395 What:           /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats
 396                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/turbo_stat
 397                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/sub_turbo_stat
 398                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/unthrottle
 399                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/powercap
 400                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/overtemp
 401                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/supply_fault
 402                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/overcurrent
 403                 /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats/occ_reset
 404 Date:           March 2016
 405 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 406                 Linux for PowerPC mailing list <linuxppc-dev@ozlabs.org>
 407 Description:    POWERNV CPUFreq driver's frequency throttle stats directory and
 408                 attributes
 409
 410                 'cpuX/cpufreq/throttle_stats' directory contains the CPU frequency
 411                 throttle stat attributes for the chip. The throttle stats of a cpu
 412                 is common across all the cpus belonging to a chip. Below are the
 413                 throttle attributes exported in the 'throttle_stats' directory:
 414
 415                 - turbo_stat : This file gives the total number of times the max
 416                 frequency is throttled to lower frequency in turbo (at and above
 417                 nominal frequency) range of frequencies.
 418
 419                 - sub_turbo_stat : This file gives the total number of times the
 420                 max frequency is throttled to lower frequency in sub-turbo(below
 421                 nominal frequency) range of frequencies.
 422
 423                 - unthrottle : This file gives the total number of times the max
 424                 frequency is unthrottled after being throttled.
 425
 426                 - powercap : This file gives the total number of times the max
 427                 frequency is throttled due to 'Power Capping'.
 428
 429                 - overtemp : This file gives the total number of times the max
 430                 frequency is throttled due to 'CPU Over Temperature'.
 431
 432                 - supply_fault : This file gives the total number of times the
 433                 max frequency is throttled due to 'Power Supply Failure'.
 434
 435                 - overcurrent : This file gives the total number of times the
 436                 max frequency is throttled due to 'Overcurrent'.
 437
 438                 - occ_reset : This file gives the total number of times the max
 439                 frequency is throttled due to 'OCC Reset'.
 440
 441                 The sysfs attributes representing different throttle reasons like
 442                 powercap, overtemp, supply_fault, overcurrent and occ_reset map to
 443                 the reasons provided by OCC firmware for throttling the frequency.
 444
 445 What:           /sys/devices/system/cpu/cpufreq/policyX/throttle_stats
 446                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/turbo_stat
 447                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/sub_turbo_stat
 448                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/unthrottle
 449                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/powercap
 450                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/overtemp
 451                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/supply_fault
 452                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/overcurrent
 453                 /sys/devices/system/cpu/cpufreq/policyX/throttle_stats/occ_reset
 454 Date:           March 2016
 455 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 456                 Linux for PowerPC mailing list <linuxppc-dev@ozlabs.org>
 457 Description:    POWERNV CPUFreq driver's frequency throttle stats directory and
 458                 attributes
 459
 460                 'policyX/throttle_stats' directory and all the attributes are same as
 461                 the /sys/devices/system/cpu/cpuX/cpufreq/throttle_stats directory and
 462                 attributes which give the frequency throttle information of the chip.
 463
 464 What:           /sys/devices/system/cpu/cpuX/regs/
 465                 /sys/devices/system/cpu/cpuX/regs/identification/
 466                 /sys/devices/system/cpu/cpuX/regs/identification/midr_el1
 467                 /sys/devices/system/cpu/cpuX/regs/identification/revidr_el1
 468 Date:           June 2016
 469 Contact:        Linux ARM Kernel Mailing list <linux-arm-kernel@lists.infradead.org>
 470 Description:    AArch64 CPU registers
 471                 'identification' directory exposes the CPU ID registers for
 472                  identifying model and revision of the CPU.
 473
 474 What:           /sys/devices/system/cpu/cpu#/cpu_capacity
 475 Date:           December 2016
 476 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 477 Description:    information about CPUs heterogeneity.
 478
 479                 cpu_capacity: capacity of cpu#.
 480
 481 What:           /sys/devices/system/cpu/vulnerabilities
 482                 /sys/devices/system/cpu/vulnerabilities/meltdown
 483                 /sys/devices/system/cpu/vulnerabilities/spectre_v1
 484                 /sys/devices/system/cpu/vulnerabilities/spectre_v2
 485                 /sys/devices/system/cpu/vulnerabilities/spec_store_bypass
 486                 /sys/devices/system/cpu/vulnerabilities/l1tf
 487 Date:           January 2018
 488 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 489 Description:    Information about CPU vulnerabilities
 490
 491                 The files are named after the code names of CPU
 492                 vulnerabilities. The output of those files reflects the
 493                 state of the CPUs in the system. Possible output values:
 494
 495                 "Not affected"    CPU is not affected by the vulnerability
 496                 "Vulnerable"      CPU is affected and no mitigation in effect
 497                 "Mitigation: $M"  CPU is affected and mitigation $M is in effect
 498
 499                 Details about the l1tf file can be found in
 500                 Documentation/admin-guide/l1tf.rst
 501
 502 What:           /sys/devices/system/cpu/smt
 503                 /sys/devices/system/cpu/smt/active
 504                 /sys/devices/system/cpu/smt/control
 505 Date:           June 2018
 506 Contact:        Linux kernel mailing list <linux-kernel@vger.kernel.org>
 507 Description:    Control Symetric Multi Threading (SMT)
 508
 509                 active:  Tells whether SMT is active (enabled and siblings online)
 510
 511                 control: Read/write interface to control SMT. Possible
 512                          values:
 513
 514                          "on"             SMT is enabled
 515                          "off"            SMT is disabled
 516                          "forceoff"       SMT is force disabled. Cannot be changed.
 517                          "notsupported"   SMT is not supported by the CPU
 518                          "notimplemented" SMT runtime toggling is not
 519                                           implemented for the architecture
 520
 521                          If control status is "forceoff" or "notsupported" writes
 522                          are rejected.
 523
 524 What:           /sys/devices/system/cpu/cpu#/power/energy_perf_bias
 525 Date:           March 2019
 526 Contact:        linux-pm@vger.kernel.org
 527 Description:    Intel Energy and Performance Bias Hint (EPB)
 528
 529                 EPB for the given CPU in a sliding scale 0 - 15, where a value
 530                 of 0 corresponds to a hint preference for highest performance
 531                 and a value of 15 corresponds to the maximum energy savings.
 532
 533                 In order to change the EPB value for the CPU, write either
 534                 a number in the 0 - 15 sliding scale above, or one of the
 535                 strings: "performance", "balance-performance", "normal",
 536                 "balance-power", "power" (that represent values reflected by
 537                 their meaning), to this attribute.
 538
 539                 This attribute is present for all online CPUs supporting the
 540                 Intel EPB feature.