Loading...
1===================================
2Documentation for /proc/sys/kernel/
3===================================
4
5.. See scripts/check-sysctl-docs to keep this up to date
6
7
8Copyright (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
9
10Copyright (c) 2009, Shen Feng<shen@cn.fujitsu.com>
11
12For general info and legal blurb, please look in
13Documentation/admin-guide/sysctl/index.rst.
14
15------------------------------------------------------------------------------
16
17This file contains documentation for the sysctl files in
18``/proc/sys/kernel/``.
19
20The files in this directory can be used to tune and monitor
21miscellaneous and general things in the operation of the Linux
22kernel. Since some of the files *can* be used to screw up your
23system, it is advisable to read both documentation and source
24before actually making adjustments.
25
26Currently, these files might (depending on your configuration)
27show up in ``/proc/sys/kernel``:
28
29.. contents:: :local:
30
31
32acct
33====
34
35::
36
37 highwater lowwater frequency
38
39If BSD-style process accounting is enabled these values control
40its behaviour. If free space on filesystem where the log lives
41goes below ``lowwater``\ % accounting suspends. If free space gets
42above ``highwater``\ % accounting resumes. ``frequency`` determines
43how often do we check the amount of free space (value is in
44seconds). Default:
45
46::
47
48 4 2 30
49
50That is, suspend accounting if free space drops below 2%; resume it
51if it increases to at least 4%; consider information about amount of
52free space valid for 30 seconds.
53
54
55acpi_video_flags
56================
57
58See Documentation/power/video.rst. This allows the video resume mode to be set,
59in a similar fashion to the ``acpi_sleep`` kernel parameter, by
60combining the following values:
61
62= =======
631 s3_bios
642 s3_mode
654 s3_beep
66= =======
67
68arch
69====
70
71The machine hardware name, the same output as ``uname -m``
72(e.g. ``x86_64`` or ``aarch64``).
73
74auto_msgmni
75===========
76
77This variable has no effect and may be removed in future kernel
78releases. Reading it always returns 0.
79Up to Linux 3.17, it enabled/disabled automatic recomputing of
80`msgmni`_
81upon memory add/remove or upon IPC namespace creation/removal.
82Echoing "1" into this file enabled msgmni automatic recomputing.
83Echoing "0" turned it off. The default value was 1.
84
85
86bootloader_type (x86 only)
87==========================
88
89This gives the bootloader type number as indicated by the bootloader,
90shifted left by 4, and OR'd with the low four bits of the bootloader
91version. The reason for this encoding is that this used to match the
92``type_of_loader`` field in the kernel header; the encoding is kept for
93backwards compatibility. That is, if the full bootloader type number
94is 0x15 and the full version number is 0x234, this file will contain
95the value 340 = 0x154.
96
97See the ``type_of_loader`` and ``ext_loader_type`` fields in
98Documentation/arch/x86/boot.rst for additional information.
99
100
101bootloader_version (x86 only)
102=============================
103
104The complete bootloader version number. In the example above, this
105file will contain the value 564 = 0x234.
106
107See the ``type_of_loader`` and ``ext_loader_ver`` fields in
108Documentation/arch/x86/boot.rst for additional information.
109
110
111bpf_stats_enabled
112=================
113
114Controls whether the kernel should collect statistics on BPF programs
115(total time spent running, number of times run...). Enabling
116statistics causes a slight reduction in performance on each program
117run. The statistics can be seen using ``bpftool``.
118
119= ===================================
1200 Don't collect statistics (default).
1211 Collect statistics.
122= ===================================
123
124
125cad_pid
126=======
127
128This is the pid which will be signalled on reboot (notably, by
129Ctrl-Alt-Delete). Writing a value to this file which doesn't
130correspond to a running process will result in ``-ESRCH``.
131
132See also `ctrl-alt-del`_.
133
134
135cap_last_cap
136============
137
138Highest valid capability of the running kernel. Exports
139``CAP_LAST_CAP`` from the kernel.
140
141
142.. _core_pattern:
143
144core_pattern
145============
146
147``core_pattern`` is used to specify a core dumpfile pattern name.
148
149* max length 127 characters; default value is "core"
150* ``core_pattern`` is used as a pattern template for the output
151 filename; certain string patterns (beginning with '%') are
152 substituted with their actual values.
153* backward compatibility with ``core_uses_pid``:
154
155 If ``core_pattern`` does not include "%p" (default does not)
156 and ``core_uses_pid`` is set, then .PID will be appended to
157 the filename.
158
159* corename format specifiers
160
161 ======== ==========================================
162 %<NUL> '%' is dropped
163 %% output one '%'
164 %p pid
165 %P global pid (init PID namespace)
166 %i tid
167 %I global tid (init PID namespace)
168 %u uid (in initial user namespace)
169 %g gid (in initial user namespace)
170 %d dump mode, matches ``PR_SET_DUMPABLE`` and
171 ``/proc/sys/fs/suid_dumpable``
172 %s signal number
173 %t UNIX time of dump
174 %h hostname
175 %e executable filename (may be shortened, could be changed by prctl etc)
176 %f executable filename
177 %E executable path
178 %c maximum size of core file by resource limit RLIMIT_CORE
179 %C CPU the task ran on
180 %<OTHER> both are dropped
181 ======== ==========================================
182
183* If the first character of the pattern is a '|', the kernel will treat
184 the rest of the pattern as a command to run. The core dump will be
185 written to the standard input of that program instead of to a file.
186
187
188core_pipe_limit
189===============
190
191This sysctl is only applicable when `core_pattern`_ is configured to
192pipe core files to a user space helper (when the first character of
193``core_pattern`` is a '|', see above).
194When collecting cores via a pipe to an application, it is occasionally
195useful for the collecting application to gather data about the
196crashing process from its ``/proc/pid`` directory.
197In order to do this safely, the kernel must wait for the collecting
198process to exit, so as not to remove the crashing processes proc files
199prematurely.
200This in turn creates the possibility that a misbehaving userspace
201collecting process can block the reaping of a crashed process simply
202by never exiting.
203This sysctl defends against that.
204It defines how many concurrent crashing processes may be piped to user
205space applications in parallel.
206If this value is exceeded, then those crashing processes above that
207value are noted via the kernel log and their cores are skipped.
2080 is a special value, indicating that unlimited processes may be
209captured in parallel, but that no waiting will take place (i.e. the
210collecting process is not guaranteed access to ``/proc/<crashing
211pid>/``).
212This value defaults to 0.
213
214
215core_sort_vma
216=============
217
218The default coredump writes VMAs in address order. By setting
219``core_sort_vma`` to 1, VMAs will be written from smallest size
220to largest size. This is known to break at least elfutils, but
221can be handy when dealing with very large (and truncated)
222coredumps where the more useful debugging details are included
223in the smaller VMAs.
224
225
226core_uses_pid
227=============
228
229The default coredump filename is "core". By setting
230``core_uses_pid`` to 1, the coredump filename becomes core.PID.
231If `core_pattern`_ does not include "%p" (default does not)
232and ``core_uses_pid`` is set, then .PID will be appended to
233the filename.
234
235
236ctrl-alt-del
237============
238
239When the value in this file is 0, ctrl-alt-del is trapped and
240sent to the ``init(1)`` program to handle a graceful restart.
241When, however, the value is > 0, Linux's reaction to a Vulcan
242Nerve Pinch (tm) will be an immediate reboot, without even
243syncing its dirty buffers.
244
245Note:
246 when a program (like dosemu) has the keyboard in 'raw'
247 mode, the ctrl-alt-del is intercepted by the program before it
248 ever reaches the kernel tty layer, and it's up to the program
249 to decide what to do with it.
250
251
252dmesg_restrict
253==============
254
255This toggle indicates whether unprivileged users are prevented
256from using ``dmesg(8)`` to view messages from the kernel's log
257buffer.
258When ``dmesg_restrict`` is set to 0 there are no restrictions.
259When ``dmesg_restrict`` is set to 1, users must have
260``CAP_SYSLOG`` to use ``dmesg(8)``.
261
262The kernel config option ``CONFIG_SECURITY_DMESG_RESTRICT`` sets the
263default value of ``dmesg_restrict``.
264
265
266domainname & hostname
267=====================
268
269These files can be used to set the NIS/YP domainname and the
270hostname of your box in exactly the same way as the commands
271domainname and hostname, i.e.::
272
273 # echo "darkstar" > /proc/sys/kernel/hostname
274 # echo "mydomain" > /proc/sys/kernel/domainname
275
276has the same effect as::
277
278 # hostname "darkstar"
279 # domainname "mydomain"
280
281Note, however, that the classic darkstar.frop.org has the
282hostname "darkstar" and DNS (Internet Domain Name Server)
283domainname "frop.org", not to be confused with the NIS (Network
284Information Service) or YP (Yellow Pages) domainname. These two
285domain names are in general different. For a detailed discussion
286see the ``hostname(1)`` man page.
287
288
289firmware_config
290===============
291
292See Documentation/driver-api/firmware/fallback-mechanisms.rst.
293
294The entries in this directory allow the firmware loader helper
295fallback to be controlled:
296
297* ``force_sysfs_fallback``, when set to 1, forces the use of the
298 fallback;
299* ``ignore_sysfs_fallback``, when set to 1, ignores any fallback.
300
301
302ftrace_dump_on_oops
303===================
304
305Determines whether ``ftrace_dump()`` should be called on an oops (or
306kernel panic). This will output the contents of the ftrace buffers to
307the console. This is very useful for capturing traces that lead to
308crashes and outputting them to a serial console.
309
310======================= ===========================================
3110 Disabled (default).
3121 Dump buffers of all CPUs.
3132(orig_cpu) Dump the buffer of the CPU that triggered the
314 oops.
315<instance> Dump the specific instance buffer on all CPUs.
316<instance>=2(orig_cpu) Dump the specific instance buffer on the CPU
317 that triggered the oops.
318======================= ===========================================
319
320Multiple instance dump is also supported, and instances are separated
321by commas. If global buffer also needs to be dumped, please specify
322the dump mode (1/2/orig_cpu) first for global buffer.
323
324So for example to dump "foo" and "bar" instance buffer on all CPUs,
325user can::
326
327 echo "foo,bar" > /proc/sys/kernel/ftrace_dump_on_oops
328
329To dump global buffer and "foo" instance buffer on all
330CPUs along with the "bar" instance buffer on CPU that triggered the
331oops, user can::
332
333 echo "1,foo,bar=2" > /proc/sys/kernel/ftrace_dump_on_oops
334
335ftrace_enabled, stack_tracer_enabled
336====================================
337
338See Documentation/trace/ftrace.rst.
339
340
341hardlockup_all_cpu_backtrace
342============================
343
344This value controls the hard lockup detector behavior when a hard
345lockup condition is detected as to whether or not to gather further
346debug information. If enabled, arch-specific all-CPU stack dumping
347will be initiated.
348
349= ============================================
3500 Do nothing. This is the default behavior.
3511 On detection capture more debug information.
352= ============================================
353
354
355hardlockup_panic
356================
357
358This parameter can be used to control whether the kernel panics
359when a hard lockup is detected.
360
361= ===========================
3620 Don't panic on hard lockup.
3631 Panic on hard lockup.
364= ===========================
365
366See Documentation/admin-guide/lockup-watchdogs.rst for more information.
367This can also be set using the nmi_watchdog kernel parameter.
368
369
370hotplug
371=======
372
373Path for the hotplug policy agent.
374Default value is ``CONFIG_UEVENT_HELPER_PATH``, which in turn defaults
375to the empty string.
376
377This file only exists when ``CONFIG_UEVENT_HELPER`` is enabled. Most
378modern systems rely exclusively on the netlink-based uevent source and
379don't need this.
380
381
382hung_task_all_cpu_backtrace
383===========================
384
385If this option is set, the kernel will send an NMI to all CPUs to dump
386their backtraces when a hung task is detected. This file shows up if
387CONFIG_DETECT_HUNG_TASK and CONFIG_SMP are enabled.
388
3890: Won't show all CPUs backtraces when a hung task is detected.
390This is the default behavior.
391
3921: Will non-maskably interrupt all CPUs and dump their backtraces when
393a hung task is detected.
394
395
396hung_task_panic
397===============
398
399Controls the kernel's behavior when a hung task is detected.
400This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
401
402= =================================================
4030 Continue operation. This is the default behavior.
4041 Panic immediately.
405= =================================================
406
407
408hung_task_check_count
409=====================
410
411The upper bound on the number of tasks that are checked.
412This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
413
414
415hung_task_detect_count
416======================
417
418Indicates the total number of tasks that have been detected as hung since
419the system boot.
420
421This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
422
423
424hung_task_timeout_secs
425======================
426
427When a task in D state did not get scheduled
428for more than this value report a warning.
429This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
430
4310 means infinite timeout, no checking is done.
432
433Possible values to set are in range {0:``LONG_MAX``/``HZ``}.
434
435
436hung_task_check_interval_secs
437=============================
438
439Hung task check interval. If hung task checking is enabled
440(see `hung_task_timeout_secs`_), the check is done every
441``hung_task_check_interval_secs`` seconds.
442This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
443
4440 (default) means use ``hung_task_timeout_secs`` as checking
445interval.
446
447Possible values to set are in range {0:``LONG_MAX``/``HZ``}.
448
449
450hung_task_warnings
451==================
452
453The maximum number of warnings to report. During a check interval
454if a hung task is detected, this value is decreased by 1.
455When this value reaches 0, no more warnings will be reported.
456This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
457
458-1: report an infinite number of warnings.
459
460
461hyperv_record_panic_msg
462=======================
463
464Controls whether the panic kmsg data should be reported to Hyper-V.
465
466= =========================================================
4670 Do not report panic kmsg data.
4681 Report the panic kmsg data. This is the default behavior.
469= =========================================================
470
471
472ignore-unaligned-usertrap
473=========================
474
475On architectures where unaligned accesses cause traps, and where this
476feature is supported (``CONFIG_SYSCTL_ARCH_UNALIGN_NO_WARN``;
477currently, ``arc``, ``parisc`` and ``loongarch``), controls whether all
478unaligned traps are logged.
479
480= =============================================================
4810 Log all unaligned accesses.
4821 Only warn the first time a process traps. This is the default
483 setting.
484= =============================================================
485
486See also `unaligned-trap`_.
487
488io_uring_disabled
489=================
490
491Prevents all processes from creating new io_uring instances. Enabling this
492shrinks the kernel's attack surface.
493
494= ======================================================================
4950 All processes can create io_uring instances as normal. This is the
496 default setting.
4971 io_uring creation is disabled (io_uring_setup() will fail with
498 -EPERM) for unprivileged processes not in the io_uring_group group.
499 Existing io_uring instances can still be used. See the
500 documentation for io_uring_group for more information.
5012 io_uring creation is disabled for all processes. io_uring_setup()
502 always fails with -EPERM. Existing io_uring instances can still be
503 used.
504= ======================================================================
505
506
507io_uring_group
508==============
509
510When io_uring_disabled is set to 1, a process must either be
511privileged (CAP_SYS_ADMIN) or be in the io_uring_group group in order
512to create an io_uring instance. If io_uring_group is set to -1 (the
513default), only processes with the CAP_SYS_ADMIN capability may create
514io_uring instances.
515
516
517kexec_load_disabled
518===================
519
520A toggle indicating if the syscalls ``kexec_load`` and
521``kexec_file_load`` have been disabled.
522This value defaults to 0 (false: ``kexec_*load`` enabled), but can be
523set to 1 (true: ``kexec_*load`` disabled).
524Once true, kexec can no longer be used, and the toggle cannot be set
525back to false.
526This allows a kexec image to be loaded before disabling the syscall,
527allowing a system to set up (and later use) an image without it being
528altered.
529Generally used together with the `modules_disabled`_ sysctl.
530
531kexec_load_limit_panic
532======================
533
534This parameter specifies a limit to the number of times the syscalls
535``kexec_load`` and ``kexec_file_load`` can be called with a crash
536image. It can only be set with a more restrictive value than the
537current one.
538
539== ======================================================
540-1 Unlimited calls to kexec. This is the default setting.
541N Number of calls left.
542== ======================================================
543
544kexec_load_limit_reboot
545=======================
546
547Similar functionality as ``kexec_load_limit_panic``, but for a normal
548image.
549
550kptr_restrict
551=============
552
553This toggle indicates whether restrictions are placed on
554exposing kernel addresses via ``/proc`` and other interfaces.
555
556When ``kptr_restrict`` is set to 0 (the default) the address is hashed
557before printing.
558(This is the equivalent to %p.)
559
560When ``kptr_restrict`` is set to 1, kernel pointers printed using the
561%pK format specifier will be replaced with 0s unless the user has
562``CAP_SYSLOG`` and effective user and group ids are equal to the real
563ids.
564This is because %pK checks are done at read() time rather than open()
565time, so if permissions are elevated between the open() and the read()
566(e.g via a setuid binary) then %pK will not leak kernel pointers to
567unprivileged users.
568Note, this is a temporary solution only.
569The correct long-term solution is to do the permission checks at
570open() time.
571Consider removing world read permissions from files that use %pK, and
572using `dmesg_restrict`_ to protect against uses of %pK in ``dmesg(8)``
573if leaking kernel pointer values to unprivileged users is a concern.
574
575When ``kptr_restrict`` is set to 2, kernel pointers printed using
576%pK will be replaced with 0s regardless of privileges.
577
578
579modprobe
580========
581
582The full path to the usermode helper for autoloading kernel modules,
583by default ``CONFIG_MODPROBE_PATH``, which in turn defaults to
584"/sbin/modprobe". This binary is executed when the kernel requests a
585module. For example, if userspace passes an unknown filesystem type
586to mount(), then the kernel will automatically request the
587corresponding filesystem module by executing this usermode helper.
588This usermode helper should insert the needed module into the kernel.
589
590This sysctl only affects module autoloading. It has no effect on the
591ability to explicitly insert modules.
592
593This sysctl can be used to debug module loading requests::
594
595 echo '#! /bin/sh' > /tmp/modprobe
596 echo 'echo "$@" >> /tmp/modprobe.log' >> /tmp/modprobe
597 echo 'exec /sbin/modprobe "$@"' >> /tmp/modprobe
598 chmod a+x /tmp/modprobe
599 echo /tmp/modprobe > /proc/sys/kernel/modprobe
600
601Alternatively, if this sysctl is set to the empty string, then module
602autoloading is completely disabled. The kernel will not try to
603execute a usermode helper at all, nor will it call the
604kernel_module_request LSM hook.
605
606If CONFIG_STATIC_USERMODEHELPER=y is set in the kernel configuration,
607then the configured static usermode helper overrides this sysctl,
608except that the empty string is still accepted to completely disable
609module autoloading as described above.
610
611modules_disabled
612================
613
614A toggle value indicating if modules are allowed to be loaded
615in an otherwise modular kernel. This toggle defaults to off
616(0), but can be set true (1). Once true, modules can be
617neither loaded nor unloaded, and the toggle cannot be set back
618to false. Generally used with the `kexec_load_disabled`_ toggle.
619
620
621.. _msgmni:
622
623msgmax, msgmnb, and msgmni
624==========================
625
626``msgmax`` is the maximum size of an IPC message, in bytes. 8192 by
627default (``MSGMAX``).
628
629``msgmnb`` is the maximum size of an IPC queue, in bytes. 16384 by
630default (``MSGMNB``).
631
632``msgmni`` is the maximum number of IPC queues. 32000 by default
633(``MSGMNI``).
634
635All of these parameters are set per ipc namespace. The maximum number of bytes
636in POSIX message queues is limited by ``RLIMIT_MSGQUEUE``. This limit is
637respected hierarchically in the each user namespace.
638
639msg_next_id, sem_next_id, and shm_next_id (System V IPC)
640========================================================
641
642These three toggles allows to specify desired id for next allocated IPC
643object: message, semaphore or shared memory respectively.
644
645By default they are equal to -1, which means generic allocation logic.
646Possible values to set are in range {0:``INT_MAX``}.
647
648Notes:
649 1) kernel doesn't guarantee, that new object will have desired id. So,
650 it's up to userspace, how to handle an object with "wrong" id.
651 2) Toggle with non-default value will be set back to -1 by kernel after
652 successful IPC object allocation. If an IPC object allocation syscall
653 fails, it is undefined if the value remains unmodified or is reset to -1.
654
655
656ngroups_max
657===========
658
659Maximum number of supplementary groups, _i.e._ the maximum size which
660``setgroups`` will accept. Exports ``NGROUPS_MAX`` from the kernel.
661
662
663
664nmi_watchdog
665============
666
667This parameter can be used to control the NMI watchdog
668(i.e. the hard lockup detector) on x86 systems.
669
670= =================================
6710 Disable the hard lockup detector.
6721 Enable the hard lockup detector.
673= =================================
674
675The hard lockup detector monitors each CPU for its ability to respond to
676timer interrupts. The mechanism utilizes CPU performance counter registers
677that are programmed to generate Non-Maskable Interrupts (NMIs) periodically
678while a CPU is busy. Hence, the alternative name 'NMI watchdog'.
679
680The NMI watchdog is disabled by default if the kernel is running as a guest
681in a KVM virtual machine. This default can be overridden by adding::
682
683 nmi_watchdog=1
684
685to the guest kernel command line (see
686Documentation/admin-guide/kernel-parameters.rst).
687
688
689nmi_wd_lpm_factor (PPC only)
690============================
691
692Factor to apply to the NMI watchdog timeout (only when ``nmi_watchdog`` is
693set to 1). This factor represents the percentage added to
694``watchdog_thresh`` when calculating the NMI watchdog timeout during an
695LPM. The soft lockup timeout is not impacted.
696
697A value of 0 means no change. The default value is 200 meaning the NMI
698watchdog is set to 30s (based on ``watchdog_thresh`` equal to 10).
699
700
701numa_balancing
702==============
703
704Enables/disables and configures automatic page fault based NUMA memory
705balancing. Memory is moved automatically to nodes that access it often.
706The value to set can be the result of ORing the following:
707
708= =================================
7090 NUMA_BALANCING_DISABLED
7101 NUMA_BALANCING_NORMAL
7112 NUMA_BALANCING_MEMORY_TIERING
712= =================================
713
714Or NUMA_BALANCING_NORMAL to optimize page placement among different
715NUMA nodes to reduce remote accessing. On NUMA machines, there is a
716performance penalty if remote memory is accessed by a CPU. When this
717feature is enabled the kernel samples what task thread is accessing
718memory by periodically unmapping pages and later trapping a page
719fault. At the time of the page fault, it is determined if the data
720being accessed should be migrated to a local memory node.
721
722The unmapping of pages and trapping faults incur additional overhead that
723ideally is offset by improved memory locality but there is no universal
724guarantee. If the target workload is already bound to NUMA nodes then this
725feature should be disabled.
726
727Or NUMA_BALANCING_MEMORY_TIERING to optimize page placement among
728different types of memory (represented as different NUMA nodes) to
729place the hot pages in the fast memory. This is implemented based on
730unmapping and page fault too.
731
732numa_balancing_promote_rate_limit_MBps
733======================================
734
735Too high promotion/demotion throughput between different memory types
736may hurt application latency. This can be used to rate limit the
737promotion throughput. The per-node max promotion throughput in MB/s
738will be limited to be no more than the set value.
739
740A rule of thumb is to set this to less than 1/10 of the PMEM node
741write bandwidth.
742
743oops_all_cpu_backtrace
744======================
745
746If this option is set, the kernel will send an NMI to all CPUs to dump
747their backtraces when an oops event occurs. It should be used as a last
748resort in case a panic cannot be triggered (to protect VMs running, for
749example) or kdump can't be collected. This file shows up if CONFIG_SMP
750is enabled.
751
7520: Won't show all CPUs backtraces when an oops is detected.
753This is the default behavior.
754
7551: Will non-maskably interrupt all CPUs and dump their backtraces when
756an oops event is detected.
757
758
759oops_limit
760==========
761
762Number of kernel oopses after which the kernel should panic when
763``panic_on_oops`` is not set. Setting this to 0 disables checking
764the count. Setting this to 1 has the same effect as setting
765``panic_on_oops=1``. The default value is 10000.
766
767
768osrelease, ostype & version
769===========================
770
771::
772
773 # cat osrelease
774 2.1.88
775 # cat ostype
776 Linux
777 # cat version
778 #5 Wed Feb 25 21:49:24 MET 1998
779
780The files ``osrelease`` and ``ostype`` should be clear enough.
781``version``
782needs a little more clarification however. The '#5' means that
783this is the fifth kernel built from this source base and the
784date behind it indicates the time the kernel was built.
785The only way to tune these values is to rebuild the kernel :-)
786
787
788overflowgid & overflowuid
789=========================
790
791if your architecture did not always support 32-bit UIDs (i.e. arm,
792i386, m68k, sh, and sparc32), a fixed UID and GID will be returned to
793applications that use the old 16-bit UID/GID system calls, if the
794actual UID or GID would exceed 65535.
795
796These sysctls allow you to change the value of the fixed UID and GID.
797The default is 65534.
798
799
800panic
801=====
802
803The value in this file determines the behaviour of the kernel on a
804panic:
805
806* if zero, the kernel will loop forever;
807* if negative, the kernel will reboot immediately;
808* if positive, the kernel will reboot after the corresponding number
809 of seconds.
810
811When you use the software watchdog, the recommended setting is 60.
812
813
814panic_on_io_nmi
815===============
816
817Controls the kernel's behavior when a CPU receives an NMI caused by
818an IO error.
819
820= ==================================================================
8210 Try to continue operation (default).
8221 Panic immediately. The IO error triggered an NMI. This indicates a
823 serious system condition which could result in IO data corruption.
824 Rather than continuing, panicking might be a better choice. Some
825 servers issue this sort of NMI when the dump button is pushed,
826 and you can use this option to take a crash dump.
827= ==================================================================
828
829
830panic_on_oops
831=============
832
833Controls the kernel's behaviour when an oops or BUG is encountered.
834
835= ===================================================================
8360 Try to continue operation.
8371 Panic immediately. If the `panic` sysctl is also non-zero then the
838 machine will be rebooted.
839= ===================================================================
840
841
842panic_on_stackoverflow
843======================
844
845Controls the kernel's behavior when detecting the overflows of
846kernel, IRQ and exception stacks except a user stack.
847This file shows up if ``CONFIG_DEBUG_STACKOVERFLOW`` is enabled.
848
849= ==========================
8500 Try to continue operation.
8511 Panic immediately.
852= ==========================
853
854
855panic_on_unrecovered_nmi
856========================
857
858The default Linux behaviour on an NMI of either memory or unknown is
859to continue operation. For many environments such as scientific
860computing it is preferable that the box is taken out and the error
861dealt with than an uncorrected parity/ECC error get propagated.
862
863A small number of systems do generate NMIs for bizarre random reasons
864such as power management so the default is off. That sysctl works like
865the existing panic controls already in that directory.
866
867
868panic_on_warn
869=============
870
871Calls panic() in the WARN() path when set to 1. This is useful to avoid
872a kernel rebuild when attempting to kdump at the location of a WARN().
873
874= ================================================
8750 Only WARN(), default behaviour.
8761 Call panic() after printing out WARN() location.
877= ================================================
878
879
880panic_print
881===========
882
883Bitmask for printing system info when panic happens. User can chose
884combination of the following bits:
885
886===== ============================================
887bit 0 print all tasks info
888bit 1 print system memory info
889bit 2 print timer info
890bit 3 print locks info if ``CONFIG_LOCKDEP`` is on
891bit 4 print ftrace buffer
892bit 5 print all printk messages in buffer
893bit 6 print all CPUs backtrace (if available in the arch)
894bit 7 print only tasks in uninterruptible (blocked) state
895===== ============================================
896
897So for example to print tasks and memory info on panic, user can::
898
899 echo 3 > /proc/sys/kernel/panic_print
900
901
902panic_on_rcu_stall
903==================
904
905When set to 1, calls panic() after RCU stall detection messages. This
906is useful to define the root cause of RCU stalls using a vmcore.
907
908= ============================================================
9090 Do not panic() when RCU stall takes place, default behavior.
9101 panic() after printing RCU stall messages.
911= ============================================================
912
913max_rcu_stall_to_panic
914======================
915
916When ``panic_on_rcu_stall`` is set to 1, this value determines the
917number of times that RCU can stall before panic() is called.
918
919When ``panic_on_rcu_stall`` is set to 0, this value is has no effect.
920
921perf_cpu_time_max_percent
922=========================
923
924Hints to the kernel how much CPU time it should be allowed to
925use to handle perf sampling events. If the perf subsystem
926is informed that its samples are exceeding this limit, it
927will drop its sampling frequency to attempt to reduce its CPU
928usage.
929
930Some perf sampling happens in NMIs. If these samples
931unexpectedly take too long to execute, the NMIs can become
932stacked up next to each other so much that nothing else is
933allowed to execute.
934
935===== ========================================================
9360 Disable the mechanism. Do not monitor or correct perf's
937 sampling rate no matter how CPU time it takes.
938
9391-100 Attempt to throttle perf's sample rate to this
940 percentage of CPU. Note: the kernel calculates an
941 "expected" length of each sample event. 100 here means
942 100% of that expected length. Even if this is set to
943 100, you may still see sample throttling if this
944 length is exceeded. Set to 0 if you truly do not care
945 how much CPU is consumed.
946===== ========================================================
947
948
949perf_event_paranoid
950===================
951
952Controls use of the performance events system by unprivileged
953users (without CAP_PERFMON). The default value is 2.
954
955For backward compatibility reasons access to system performance
956monitoring and observability remains open for CAP_SYS_ADMIN
957privileged processes but CAP_SYS_ADMIN usage for secure system
958performance monitoring and observability operations is discouraged
959with respect to CAP_PERFMON use cases.
960
961=== ==================================================================
962 -1 Allow use of (almost) all events by all users.
963
964 Ignore mlock limit after perf_event_mlock_kb without
965 ``CAP_IPC_LOCK``.
966
967>=0 Disallow ftrace function tracepoint by users without
968 ``CAP_PERFMON``.
969
970 Disallow raw tracepoint access by users without ``CAP_PERFMON``.
971
972>=1 Disallow CPU event access by users without ``CAP_PERFMON``.
973
974>=2 Disallow kernel profiling by users without ``CAP_PERFMON``.
975=== ==================================================================
976
977
978perf_event_max_stack
979====================
980
981Controls maximum number of stack frames to copy for (``attr.sample_type &
982PERF_SAMPLE_CALLCHAIN``) configured events, for instance, when using
983'``perf record -g``' or '``perf trace --call-graph fp``'.
984
985This can only be done when no events are in use that have callchains
986enabled, otherwise writing to this file will return ``-EBUSY``.
987
988The default value is 127.
989
990
991perf_event_mlock_kb
992===================
993
994Control size of per-cpu ring buffer not counted against mlock limit.
995
996The default value is 512 + 1 page
997
998
999perf_event_max_contexts_per_stack
1000=================================
1001
1002Controls maximum number of stack frame context entries for
1003(``attr.sample_type & PERF_SAMPLE_CALLCHAIN``) configured events, for
1004instance, when using '``perf record -g``' or '``perf trace --call-graph fp``'.
1005
1006This can only be done when no events are in use that have callchains
1007enabled, otherwise writing to this file will return ``-EBUSY``.
1008
1009The default value is 8.
1010
1011
1012perf_user_access (arm64 and riscv only)
1013=======================================
1014
1015Controls user space access for reading perf event counters.
1016
1017arm64
1018=====
1019
1020The default value is 0 (access disabled).
1021
1022When set to 1, user space can read performance monitor counter registers
1023directly.
1024
1025See Documentation/arch/arm64/perf.rst for more information.
1026
1027riscv
1028=====
1029
1030When set to 0, user space access is disabled.
1031
1032The default value is 1, user space can read performance monitor counter
1033registers through perf, any direct access without perf intervention will trigger
1034an illegal instruction.
1035
1036When set to 2, which enables legacy mode (user space has direct access to cycle
1037and insret CSRs only). Note that this legacy value is deprecated and will be
1038removed once all user space applications are fixed.
1039
1040Note that the time CSR is always directly accessible to all modes.
1041
1042pid_max
1043=======
1044
1045PID allocation wrap value. When the kernel's next PID value
1046reaches this value, it wraps back to a minimum PID value.
1047PIDs of value ``pid_max`` or larger are not allocated.
1048
1049
1050ns_last_pid
1051===========
1052
1053The last pid allocated in the current (the one task using this sysctl
1054lives in) pid namespace. When selecting a pid for a next task on fork
1055kernel tries to allocate a number starting from this one.
1056
1057
1058powersave-nap (PPC only)
1059========================
1060
1061If set, Linux-PPC will use the 'nap' mode of powersaving,
1062otherwise the 'doze' mode will be used.
1063
1064
1065==============================================================
1066
1067printk
1068======
1069
1070The four values in printk denote: ``console_loglevel``,
1071``default_message_loglevel``, ``minimum_console_loglevel`` and
1072``default_console_loglevel`` respectively.
1073
1074These values influence printk() behavior when printing or
1075logging error messages. See '``man 2 syslog``' for more info on
1076the different loglevels.
1077
1078======================== =====================================
1079console_loglevel messages with a higher priority than
1080 this will be printed to the console
1081default_message_loglevel messages without an explicit priority
1082 will be printed with this priority
1083minimum_console_loglevel minimum (highest) value to which
1084 console_loglevel can be set
1085default_console_loglevel default value for console_loglevel
1086======================== =====================================
1087
1088
1089printk_delay
1090============
1091
1092Delay each printk message in ``printk_delay`` milliseconds
1093
1094Value from 0 - 10000 is allowed.
1095
1096
1097printk_ratelimit
1098================
1099
1100Some warning messages are rate limited. ``printk_ratelimit`` specifies
1101the minimum length of time between these messages (in seconds).
1102The default value is 5 seconds.
1103
1104A value of 0 will disable rate limiting.
1105
1106
1107printk_ratelimit_burst
1108======================
1109
1110While long term we enforce one message per `printk_ratelimit`_
1111seconds, we do allow a burst of messages to pass through.
1112``printk_ratelimit_burst`` specifies the number of messages we can
1113send before ratelimiting kicks in.
1114
1115The default value is 10 messages.
1116
1117
1118printk_devkmsg
1119==============
1120
1121Control the logging to ``/dev/kmsg`` from userspace:
1122
1123========= =============================================
1124ratelimit default, ratelimited
1125on unlimited logging to /dev/kmsg from userspace
1126off logging to /dev/kmsg disabled
1127========= =============================================
1128
1129The kernel command line parameter ``printk.devkmsg=`` overrides this and is
1130a one-time setting until next reboot: once set, it cannot be changed by
1131this sysctl interface anymore.
1132
1133==============================================================
1134
1135
1136pty
1137===
1138
1139See Documentation/filesystems/devpts.rst.
1140
1141
1142random
1143======
1144
1145This is a directory, with the following entries:
1146
1147* ``boot_id``: a UUID generated the first time this is retrieved, and
1148 unvarying after that;
1149
1150* ``uuid``: a UUID generated every time this is retrieved (this can
1151 thus be used to generate UUIDs at will);
1152
1153* ``entropy_avail``: the pool's entropy count, in bits;
1154
1155* ``poolsize``: the entropy pool size, in bits;
1156
1157* ``urandom_min_reseed_secs``: obsolete (used to determine the minimum
1158 number of seconds between urandom pool reseeding). This file is
1159 writable for compatibility purposes, but writing to it has no effect
1160 on any RNG behavior;
1161
1162* ``write_wakeup_threshold``: when the entropy count drops below this
1163 (as a number of bits), processes waiting to write to ``/dev/random``
1164 are woken up. This file is writable for compatibility purposes, but
1165 writing to it has no effect on any RNG behavior.
1166
1167
1168randomize_va_space
1169==================
1170
1171This option can be used to select the type of process address
1172space randomization that is used in the system, for architectures
1173that support this feature.
1174
1175== ===========================================================================
11760 Turn the process address space randomization off. This is the
1177 default for architectures that do not support this feature anyways,
1178 and kernels that are booted with the "norandmaps" parameter.
1179
11801 Make the addresses of mmap base, stack and VDSO page randomized.
1181 This, among other things, implies that shared libraries will be
1182 loaded to random addresses. Also for PIE-linked binaries, the
1183 location of code start is randomized. This is the default if the
1184 ``CONFIG_COMPAT_BRK`` option is enabled.
1185
11862 Additionally enable heap randomization. This is the default if
1187 ``CONFIG_COMPAT_BRK`` is disabled.
1188
1189 There are a few legacy applications out there (such as some ancient
1190 versions of libc.so.5 from 1996) that assume that brk area starts
1191 just after the end of the code+bss. These applications break when
1192 start of the brk area is randomized. There are however no known
1193 non-legacy applications that would be broken this way, so for most
1194 systems it is safe to choose full randomization.
1195
1196 Systems with ancient and/or broken binaries should be configured
1197 with ``CONFIG_COMPAT_BRK`` enabled, which excludes the heap from process
1198 address space randomization.
1199== ===========================================================================
1200
1201
1202real-root-dev
1203=============
1204
1205See Documentation/admin-guide/initrd.rst.
1206
1207
1208reboot-cmd (SPARC only)
1209=======================
1210
1211??? This seems to be a way to give an argument to the Sparc
1212ROM/Flash boot loader. Maybe to tell it what to do after
1213rebooting. ???
1214
1215
1216sched_energy_aware
1217==================
1218
1219Enables/disables Energy Aware Scheduling (EAS). EAS starts
1220automatically on platforms where it can run (that is,
1221platforms with asymmetric CPU topologies and having an Energy
1222Model available). If your platform happens to meet the
1223requirements for EAS but you do not want to use it, change
1224this value to 0. On Non-EAS platforms, write operation fails and
1225read doesn't return anything.
1226
1227task_delayacct
1228===============
1229
1230Enables/disables task delay accounting (see
1231Documentation/accounting/delay-accounting.rst. Enabling this feature incurs
1232a small amount of overhead in the scheduler but is useful for debugging
1233and performance tuning. It is required by some tools such as iotop.
1234
1235sched_schedstats
1236================
1237
1238Enables/disables scheduler statistics. Enabling this feature
1239incurs a small amount of overhead in the scheduler but is
1240useful for debugging and performance tuning.
1241
1242sched_util_clamp_min
1243====================
1244
1245Max allowed *minimum* utilization.
1246
1247Default value is 1024, which is the maximum possible value.
1248
1249It means that any requested uclamp.min value cannot be greater than
1250sched_util_clamp_min, i.e., it is restricted to the range
1251[0:sched_util_clamp_min].
1252
1253sched_util_clamp_max
1254====================
1255
1256Max allowed *maximum* utilization.
1257
1258Default value is 1024, which is the maximum possible value.
1259
1260It means that any requested uclamp.max value cannot be greater than
1261sched_util_clamp_max, i.e., it is restricted to the range
1262[0:sched_util_clamp_max].
1263
1264sched_util_clamp_min_rt_default
1265===============================
1266
1267By default Linux is tuned for performance. Which means that RT tasks always run
1268at the highest frequency and most capable (highest capacity) CPU (in
1269heterogeneous systems).
1270
1271Uclamp achieves this by setting the requested uclamp.min of all RT tasks to
12721024 by default, which effectively boosts the tasks to run at the highest
1273frequency and biases them to run on the biggest CPU.
1274
1275This knob allows admins to change the default behavior when uclamp is being
1276used. In battery powered devices particularly, running at the maximum
1277capacity and frequency will increase energy consumption and shorten the battery
1278life.
1279
1280This knob is only effective for RT tasks which the user hasn't modified their
1281requested uclamp.min value via sched_setattr() syscall.
1282
1283This knob will not escape the range constraint imposed by sched_util_clamp_min
1284defined above.
1285
1286For example if
1287
1288 sched_util_clamp_min_rt_default = 800
1289 sched_util_clamp_min = 600
1290
1291Then the boost will be clamped to 600 because 800 is outside of the permissible
1292range of [0:600]. This could happen for instance if a powersave mode will
1293restrict all boosts temporarily by modifying sched_util_clamp_min. As soon as
1294this restriction is lifted, the requested sched_util_clamp_min_rt_default
1295will take effect.
1296
1297seccomp
1298=======
1299
1300See Documentation/userspace-api/seccomp_filter.rst.
1301
1302
1303sg-big-buff
1304===========
1305
1306This file shows the size of the generic SCSI (sg) buffer.
1307You can't tune it just yet, but you could change it on
1308compile time by editing ``include/scsi/sg.h`` and changing
1309the value of ``SG_BIG_BUFF``.
1310
1311There shouldn't be any reason to change this value. If
1312you can come up with one, you probably know what you
1313are doing anyway :)
1314
1315
1316shmall
1317======
1318
1319This parameter sets the total amount of shared memory pages that can be used
1320inside ipc namespace. The shared memory pages counting occurs for each ipc
1321namespace separately and is not inherited. Hence, ``shmall`` should always be at
1322least ``ceil(shmmax/PAGE_SIZE)``.
1323
1324If you are not sure what the default ``PAGE_SIZE`` is on your Linux
1325system, you can run the following command::
1326
1327 # getconf PAGE_SIZE
1328
1329To reduce or disable the ability to allocate shared memory, you must create a
1330new ipc namespace, set this parameter to the required value and prohibit the
1331creation of a new ipc namespace in the current user namespace or cgroups can
1332be used.
1333
1334shmmax
1335======
1336
1337This value can be used to query and set the run time limit
1338on the maximum shared memory segment size that can be created.
1339Shared memory segments up to 1Gb are now supported in the
1340kernel. This value defaults to ``SHMMAX``.
1341
1342
1343shmmni
1344======
1345
1346This value determines the maximum number of shared memory segments.
13474096 by default (``SHMMNI``).
1348
1349
1350shm_rmid_forced
1351===============
1352
1353Linux lets you set resource limits, including how much memory one
1354process can consume, via ``setrlimit(2)``. Unfortunately, shared memory
1355segments are allowed to exist without association with any process, and
1356thus might not be counted against any resource limits. If enabled,
1357shared memory segments are automatically destroyed when their attach
1358count becomes zero after a detach or a process termination. It will
1359also destroy segments that were created, but never attached to, on exit
1360from the process. The only use left for ``IPC_RMID`` is to immediately
1361destroy an unattached segment. Of course, this breaks the way things are
1362defined, so some applications might stop working. Note that this
1363feature will do you no good unless you also configure your resource
1364limits (in particular, ``RLIMIT_AS`` and ``RLIMIT_NPROC``). Most systems don't
1365need this.
1366
1367Note that if you change this from 0 to 1, already created segments
1368without users and with a dead originative process will be destroyed.
1369
1370
1371sysctl_writes_strict
1372====================
1373
1374Control how file position affects the behavior of updating sysctl values
1375via the ``/proc/sys`` interface:
1376
1377 == ======================================================================
1378 -1 Legacy per-write sysctl value handling, with no printk warnings.
1379 Each write syscall must fully contain the sysctl value to be
1380 written, and multiple writes on the same sysctl file descriptor
1381 will rewrite the sysctl value, regardless of file position.
1382 0 Same behavior as above, but warn about processes that perform writes
1383 to a sysctl file descriptor when the file position is not 0.
1384 1 (default) Respect file position when writing sysctl strings. Multiple
1385 writes will append to the sysctl value buffer. Anything past the max
1386 length of the sysctl value buffer will be ignored. Writes to numeric
1387 sysctl entries must always be at file position 0 and the value must
1388 be fully contained in the buffer sent in the write syscall.
1389 == ======================================================================
1390
1391
1392softlockup_all_cpu_backtrace
1393============================
1394
1395This value controls the soft lockup detector thread's behavior
1396when a soft lockup condition is detected as to whether or not
1397to gather further debug information. If enabled, each cpu will
1398be issued an NMI and instructed to capture stack trace.
1399
1400This feature is only applicable for architectures which support
1401NMI.
1402
1403= ============================================
14040 Do nothing. This is the default behavior.
14051 On detection capture more debug information.
1406= ============================================
1407
1408
1409softlockup_panic
1410=================
1411
1412This parameter can be used to control whether the kernel panics
1413when a soft lockup is detected.
1414
1415= ============================================
14160 Don't panic on soft lockup.
14171 Panic on soft lockup.
1418= ============================================
1419
1420This can also be set using the softlockup_panic kernel parameter.
1421
1422
1423soft_watchdog
1424=============
1425
1426This parameter can be used to control the soft lockup detector.
1427
1428= =================================
14290 Disable the soft lockup detector.
14301 Enable the soft lockup detector.
1431= =================================
1432
1433The soft lockup detector monitors CPUs for threads that are hogging the CPUs
1434without rescheduling voluntarily, and thus prevent the 'migration/N' threads
1435from running, causing the watchdog work fail to execute. The mechanism depends
1436on the CPUs ability to respond to timer interrupts which are needed for the
1437watchdog work to be queued by the watchdog timer function, otherwise the NMI
1438watchdog — if enabled — can detect a hard lockup condition.
1439
1440
1441split_lock_mitigate (x86 only)
1442==============================
1443
1444On x86, each "split lock" imposes a system-wide performance penalty. On larger
1445systems, large numbers of split locks from unprivileged users can result in
1446denials of service to well-behaved and potentially more important users.
1447
1448The kernel mitigates these bad users by detecting split locks and imposing
1449penalties: forcing them to wait and only allowing one core to execute split
1450locks at a time.
1451
1452These mitigations can make those bad applications unbearably slow. Setting
1453split_lock_mitigate=0 may restore some application performance, but will also
1454increase system exposure to denial of service attacks from split lock users.
1455
1456= ===================================================================
14570 Disable the mitigation mode - just warns the split lock on kernel log
1458 and exposes the system to denials of service from the split lockers.
14591 Enable the mitigation mode (this is the default) - penalizes the split
1460 lockers with intentional performance degradation.
1461= ===================================================================
1462
1463
1464stack_erasing
1465=============
1466
1467This parameter can be used to control kernel stack erasing at the end
1468of syscalls for kernels built with ``CONFIG_GCC_PLUGIN_STACKLEAK``.
1469
1470That erasing reduces the information which kernel stack leak bugs
1471can reveal and blocks some uninitialized stack variable attacks.
1472The tradeoff is the performance impact: on a single CPU system kernel
1473compilation sees a 1% slowdown, other systems and workloads may vary.
1474
1475= ====================================================================
14760 Kernel stack erasing is disabled, STACKLEAK_METRICS are not updated.
14771 Kernel stack erasing is enabled (default), it is performed before
1478 returning to the userspace at the end of syscalls.
1479= ====================================================================
1480
1481
1482stop-a (SPARC only)
1483===================
1484
1485Controls Stop-A:
1486
1487= ====================================
14880 Stop-A has no effect.
14891 Stop-A breaks to the PROM (default).
1490= ====================================
1491
1492Stop-A is always enabled on a panic, so that the user can return to
1493the boot PROM.
1494
1495
1496sysrq
1497=====
1498
1499See Documentation/admin-guide/sysrq.rst.
1500
1501
1502tainted
1503=======
1504
1505Non-zero if the kernel has been tainted. Numeric values, which can be
1506ORed together. The letters are seen in "Tainted" line of Oops reports.
1507
1508====== ===== ==============================================================
1509 1 `(P)` proprietary module was loaded
1510 2 `(F)` module was force loaded
1511 4 `(S)` kernel running on an out of specification system
1512 8 `(R)` module was force unloaded
1513 16 `(M)` processor reported a Machine Check Exception (MCE)
1514 32 `(B)` bad page referenced or some unexpected page flags
1515 64 `(U)` taint requested by userspace application
1516 128 `(D)` kernel died recently, i.e. there was an OOPS or BUG
1517 256 `(A)` an ACPI table was overridden by user
1518 512 `(W)` kernel issued warning
1519 1024 `(C)` staging driver was loaded
1520 2048 `(I)` workaround for bug in platform firmware applied
1521 4096 `(O)` externally-built ("out-of-tree") module was loaded
1522 8192 `(E)` unsigned module was loaded
1523 16384 `(L)` soft lockup occurred
1524 32768 `(K)` kernel has been live patched
1525 65536 `(X)` Auxiliary taint, defined and used by for distros
1526131072 `(T)` The kernel was built with the struct randomization plugin
1527====== ===== ==============================================================
1528
1529See Documentation/admin-guide/tainted-kernels.rst for more information.
1530
1531Note:
1532 writes to this sysctl interface will fail with ``EINVAL`` if the kernel is
1533 booted with the command line option ``panic_on_taint=<bitmask>,nousertaint``
1534 and any of the ORed together values being written to ``tainted`` match with
1535 the bitmask declared on panic_on_taint.
1536 See Documentation/admin-guide/kernel-parameters.rst for more details on
1537 that particular kernel command line option and its optional
1538 ``nousertaint`` switch.
1539
1540threads-max
1541===========
1542
1543This value controls the maximum number of threads that can be created
1544using ``fork()``.
1545
1546During initialization the kernel sets this value such that even if the
1547maximum number of threads is created, the thread structures occupy only
1548a part (1/8th) of the available RAM pages.
1549
1550The minimum value that can be written to ``threads-max`` is 1.
1551
1552The maximum value that can be written to ``threads-max`` is given by the
1553constant ``FUTEX_TID_MASK`` (0x3fffffff).
1554
1555If a value outside of this range is written to ``threads-max`` an
1556``EINVAL`` error occurs.
1557
1558
1559traceoff_on_warning
1560===================
1561
1562When set, disables tracing (see Documentation/trace/ftrace.rst) when a
1563``WARN()`` is hit.
1564
1565
1566tracepoint_printk
1567=================
1568
1569When tracepoints are sent to printk() (enabled by the ``tp_printk``
1570boot parameter), this entry provides runtime control::
1571
1572 echo 0 > /proc/sys/kernel/tracepoint_printk
1573
1574will stop tracepoints from being sent to printk(), and::
1575
1576 echo 1 > /proc/sys/kernel/tracepoint_printk
1577
1578will send them to printk() again.
1579
1580This only works if the kernel was booted with ``tp_printk`` enabled.
1581
1582See Documentation/admin-guide/kernel-parameters.rst and
1583Documentation/trace/boottime-trace.rst.
1584
1585
1586unaligned-trap
1587==============
1588
1589On architectures where unaligned accesses cause traps, and where this
1590feature is supported (``CONFIG_SYSCTL_ARCH_UNALIGN_ALLOW``; currently,
1591``arc``, ``parisc`` and ``loongarch``), controls whether unaligned traps
1592are caught and emulated (instead of failing).
1593
1594= ========================================================
15950 Do not emulate unaligned accesses.
15961 Emulate unaligned accesses. This is the default setting.
1597= ========================================================
1598
1599See also `ignore-unaligned-usertrap`_.
1600
1601
1602unknown_nmi_panic
1603=================
1604
1605The value in this file affects behavior of handling NMI. When the
1606value is non-zero, unknown NMI is trapped and then panic occurs. At
1607that time, kernel debugging information is displayed on console.
1608
1609NMI switch that most IA32 servers have fires unknown NMI up, for
1610example. If a system hangs up, try pressing the NMI switch.
1611
1612
1613unprivileged_bpf_disabled
1614=========================
1615
1616Writing 1 to this entry will disable unprivileged calls to ``bpf()``;
1617once disabled, calling ``bpf()`` without ``CAP_SYS_ADMIN`` or ``CAP_BPF``
1618will return ``-EPERM``. Once set to 1, this can't be cleared from the
1619running kernel anymore.
1620
1621Writing 2 to this entry will also disable unprivileged calls to ``bpf()``,
1622however, an admin can still change this setting later on, if needed, by
1623writing 0 or 1 to this entry.
1624
1625If ``BPF_UNPRIV_DEFAULT_OFF`` is enabled in the kernel config, then this
1626entry will default to 2 instead of 0.
1627
1628= =============================================================
16290 Unprivileged calls to ``bpf()`` are enabled
16301 Unprivileged calls to ``bpf()`` are disabled without recovery
16312 Unprivileged calls to ``bpf()`` are disabled
1632= =============================================================
1633
1634
1635warn_limit
1636==========
1637
1638Number of kernel warnings after which the kernel should panic when
1639``panic_on_warn`` is not set. Setting this to 0 disables checking
1640the warning count. Setting this to 1 has the same effect as setting
1641``panic_on_warn=1``. The default value is 0.
1642
1643
1644watchdog
1645========
1646
1647This parameter can be used to disable or enable the soft lockup detector
1648*and* the NMI watchdog (i.e. the hard lockup detector) at the same time.
1649
1650= ==============================
16510 Disable both lockup detectors.
16521 Enable both lockup detectors.
1653= ==============================
1654
1655The soft lockup detector and the NMI watchdog can also be disabled or
1656enabled individually, using the ``soft_watchdog`` and ``nmi_watchdog``
1657parameters.
1658If the ``watchdog`` parameter is read, for example by executing::
1659
1660 cat /proc/sys/kernel/watchdog
1661
1662the output of this command (0 or 1) shows the logical OR of
1663``soft_watchdog`` and ``nmi_watchdog``.
1664
1665
1666watchdog_cpumask
1667================
1668
1669This value can be used to control on which cpus the watchdog may run.
1670The default cpumask is all possible cores, but if ``NO_HZ_FULL`` is
1671enabled in the kernel config, and cores are specified with the
1672``nohz_full=`` boot argument, those cores are excluded by default.
1673Offline cores can be included in this mask, and if the core is later
1674brought online, the watchdog will be started based on the mask value.
1675
1676Typically this value would only be touched in the ``nohz_full`` case
1677to re-enable cores that by default were not running the watchdog,
1678if a kernel lockup was suspected on those cores.
1679
1680The argument value is the standard cpulist format for cpumasks,
1681so for example to enable the watchdog on cores 0, 2, 3, and 4 you
1682might say::
1683
1684 echo 0,2-4 > /proc/sys/kernel/watchdog_cpumask
1685
1686
1687watchdog_thresh
1688===============
1689
1690This value can be used to control the frequency of hrtimer and NMI
1691events and the soft and hard lockup thresholds. The default threshold
1692is 10 seconds.
1693
1694The softlockup threshold is (``2 * watchdog_thresh``). Setting this
1695tunable to zero will disable lockup detection altogether.
1===================================
2Documentation for /proc/sys/kernel/
3===================================
4
5.. See scripts/check-sysctl-docs to keep this up to date
6
7
8Copyright (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
9
10Copyright (c) 2009, Shen Feng<shen@cn.fujitsu.com>
11
12For general info and legal blurb, please look in
13Documentation/admin-guide/sysctl/index.rst.
14
15------------------------------------------------------------------------------
16
17This file contains documentation for the sysctl files in
18``/proc/sys/kernel/``.
19
20The files in this directory can be used to tune and monitor
21miscellaneous and general things in the operation of the Linux
22kernel. Since some of the files *can* be used to screw up your
23system, it is advisable to read both documentation and source
24before actually making adjustments.
25
26Currently, these files might (depending on your configuration)
27show up in ``/proc/sys/kernel``:
28
29.. contents:: :local:
30
31
32acct
33====
34
35::
36
37 highwater lowwater frequency
38
39If BSD-style process accounting is enabled these values control
40its behaviour. If free space on filesystem where the log lives
41goes below ``lowwater``\ % accounting suspends. If free space gets
42above ``highwater``\ % accounting resumes. ``frequency`` determines
43how often do we check the amount of free space (value is in
44seconds). Default:
45
46::
47
48 4 2 30
49
50That is, suspend accounting if free space drops below 2%; resume it
51if it increases to at least 4%; consider information about amount of
52free space valid for 30 seconds.
53
54
55acpi_video_flags
56================
57
58See Documentation/power/video.rst. This allows the video resume mode to be set,
59in a similar fashion to the ``acpi_sleep`` kernel parameter, by
60combining the following values:
61
62= =======
631 s3_bios
642 s3_mode
654 s3_beep
66= =======
67
68arch
69====
70
71The machine hardware name, the same output as ``uname -m``
72(e.g. ``x86_64`` or ``aarch64``).
73
74auto_msgmni
75===========
76
77This variable has no effect and may be removed in future kernel
78releases. Reading it always returns 0.
79Up to Linux 3.17, it enabled/disabled automatic recomputing of
80`msgmni`_
81upon memory add/remove or upon IPC namespace creation/removal.
82Echoing "1" into this file enabled msgmni automatic recomputing.
83Echoing "0" turned it off. The default value was 1.
84
85
86bootloader_type (x86 only)
87==========================
88
89This gives the bootloader type number as indicated by the bootloader,
90shifted left by 4, and OR'd with the low four bits of the bootloader
91version. The reason for this encoding is that this used to match the
92``type_of_loader`` field in the kernel header; the encoding is kept for
93backwards compatibility. That is, if the full bootloader type number
94is 0x15 and the full version number is 0x234, this file will contain
95the value 340 = 0x154.
96
97See the ``type_of_loader`` and ``ext_loader_type`` fields in
98Documentation/x86/boot.rst for additional information.
99
100
101bootloader_version (x86 only)
102=============================
103
104The complete bootloader version number. In the example above, this
105file will contain the value 564 = 0x234.
106
107See the ``type_of_loader`` and ``ext_loader_ver`` fields in
108Documentation/x86/boot.rst for additional information.
109
110
111bpf_stats_enabled
112=================
113
114Controls whether the kernel should collect statistics on BPF programs
115(total time spent running, number of times run...). Enabling
116statistics causes a slight reduction in performance on each program
117run. The statistics can be seen using ``bpftool``.
118
119= ===================================
1200 Don't collect statistics (default).
1211 Collect statistics.
122= ===================================
123
124
125cad_pid
126=======
127
128This is the pid which will be signalled on reboot (notably, by
129Ctrl-Alt-Delete). Writing a value to this file which doesn't
130correspond to a running process will result in ``-ESRCH``.
131
132See also `ctrl-alt-del`_.
133
134
135cap_last_cap
136============
137
138Highest valid capability of the running kernel. Exports
139``CAP_LAST_CAP`` from the kernel.
140
141
142.. _core_pattern:
143
144core_pattern
145============
146
147``core_pattern`` is used to specify a core dumpfile pattern name.
148
149* max length 127 characters; default value is "core"
150* ``core_pattern`` is used as a pattern template for the output
151 filename; certain string patterns (beginning with '%') are
152 substituted with their actual values.
153* backward compatibility with ``core_uses_pid``:
154
155 If ``core_pattern`` does not include "%p" (default does not)
156 and ``core_uses_pid`` is set, then .PID will be appended to
157 the filename.
158
159* corename format specifiers
160
161 ======== ==========================================
162 %<NUL> '%' is dropped
163 %% output one '%'
164 %p pid
165 %P global pid (init PID namespace)
166 %i tid
167 %I global tid (init PID namespace)
168 %u uid (in initial user namespace)
169 %g gid (in initial user namespace)
170 %d dump mode, matches ``PR_SET_DUMPABLE`` and
171 ``/proc/sys/fs/suid_dumpable``
172 %s signal number
173 %t UNIX time of dump
174 %h hostname
175 %e executable filename (may be shortened, could be changed by prctl etc)
176 %f executable filename
177 %E executable path
178 %c maximum size of core file by resource limit RLIMIT_CORE
179 %C CPU the task ran on
180 %<OTHER> both are dropped
181 ======== ==========================================
182
183* If the first character of the pattern is a '|', the kernel will treat
184 the rest of the pattern as a command to run. The core dump will be
185 written to the standard input of that program instead of to a file.
186
187
188core_pipe_limit
189===============
190
191This sysctl is only applicable when `core_pattern`_ is configured to
192pipe core files to a user space helper (when the first character of
193``core_pattern`` is a '|', see above).
194When collecting cores via a pipe to an application, it is occasionally
195useful for the collecting application to gather data about the
196crashing process from its ``/proc/pid`` directory.
197In order to do this safely, the kernel must wait for the collecting
198process to exit, so as not to remove the crashing processes proc files
199prematurely.
200This in turn creates the possibility that a misbehaving userspace
201collecting process can block the reaping of a crashed process simply
202by never exiting.
203This sysctl defends against that.
204It defines how many concurrent crashing processes may be piped to user
205space applications in parallel.
206If this value is exceeded, then those crashing processes above that
207value are noted via the kernel log and their cores are skipped.
2080 is a special value, indicating that unlimited processes may be
209captured in parallel, but that no waiting will take place (i.e. the
210collecting process is not guaranteed access to ``/proc/<crashing
211pid>/``).
212This value defaults to 0.
213
214
215core_uses_pid
216=============
217
218The default coredump filename is "core". By setting
219``core_uses_pid`` to 1, the coredump filename becomes core.PID.
220If `core_pattern`_ does not include "%p" (default does not)
221and ``core_uses_pid`` is set, then .PID will be appended to
222the filename.
223
224
225ctrl-alt-del
226============
227
228When the value in this file is 0, ctrl-alt-del is trapped and
229sent to the ``init(1)`` program to handle a graceful restart.
230When, however, the value is > 0, Linux's reaction to a Vulcan
231Nerve Pinch (tm) will be an immediate reboot, without even
232syncing its dirty buffers.
233
234Note:
235 when a program (like dosemu) has the keyboard in 'raw'
236 mode, the ctrl-alt-del is intercepted by the program before it
237 ever reaches the kernel tty layer, and it's up to the program
238 to decide what to do with it.
239
240
241dmesg_restrict
242==============
243
244This toggle indicates whether unprivileged users are prevented
245from using ``dmesg(8)`` to view messages from the kernel's log
246buffer.
247When ``dmesg_restrict`` is set to 0 there are no restrictions.
248When ``dmesg_restrict`` is set to 1, users must have
249``CAP_SYSLOG`` to use ``dmesg(8)``.
250
251The kernel config option ``CONFIG_SECURITY_DMESG_RESTRICT`` sets the
252default value of ``dmesg_restrict``.
253
254
255domainname & hostname
256=====================
257
258These files can be used to set the NIS/YP domainname and the
259hostname of your box in exactly the same way as the commands
260domainname and hostname, i.e.::
261
262 # echo "darkstar" > /proc/sys/kernel/hostname
263 # echo "mydomain" > /proc/sys/kernel/domainname
264
265has the same effect as::
266
267 # hostname "darkstar"
268 # domainname "mydomain"
269
270Note, however, that the classic darkstar.frop.org has the
271hostname "darkstar" and DNS (Internet Domain Name Server)
272domainname "frop.org", not to be confused with the NIS (Network
273Information Service) or YP (Yellow Pages) domainname. These two
274domain names are in general different. For a detailed discussion
275see the ``hostname(1)`` man page.
276
277
278firmware_config
279===============
280
281See Documentation/driver-api/firmware/fallback-mechanisms.rst.
282
283The entries in this directory allow the firmware loader helper
284fallback to be controlled:
285
286* ``force_sysfs_fallback``, when set to 1, forces the use of the
287 fallback;
288* ``ignore_sysfs_fallback``, when set to 1, ignores any fallback.
289
290
291ftrace_dump_on_oops
292===================
293
294Determines whether ``ftrace_dump()`` should be called on an oops (or
295kernel panic). This will output the contents of the ftrace buffers to
296the console. This is very useful for capturing traces that lead to
297crashes and outputting them to a serial console.
298
299= ===================================================
3000 Disabled (default).
3011 Dump buffers of all CPUs.
3022 Dump the buffer of the CPU that triggered the oops.
303= ===================================================
304
305
306ftrace_enabled, stack_tracer_enabled
307====================================
308
309See Documentation/trace/ftrace.rst.
310
311
312hardlockup_all_cpu_backtrace
313============================
314
315This value controls the hard lockup detector behavior when a hard
316lockup condition is detected as to whether or not to gather further
317debug information. If enabled, arch-specific all-CPU stack dumping
318will be initiated.
319
320= ============================================
3210 Do nothing. This is the default behavior.
3221 On detection capture more debug information.
323= ============================================
324
325
326hardlockup_panic
327================
328
329This parameter can be used to control whether the kernel panics
330when a hard lockup is detected.
331
332= ===========================
3330 Don't panic on hard lockup.
3341 Panic on hard lockup.
335= ===========================
336
337See Documentation/admin-guide/lockup-watchdogs.rst for more information.
338This can also be set using the nmi_watchdog kernel parameter.
339
340
341hotplug
342=======
343
344Path for the hotplug policy agent.
345Default value is ``CONFIG_UEVENT_HELPER_PATH``, which in turn defaults
346to the empty string.
347
348This file only exists when ``CONFIG_UEVENT_HELPER`` is enabled. Most
349modern systems rely exclusively on the netlink-based uevent source and
350don't need this.
351
352
353hung_task_all_cpu_backtrace
354===========================
355
356If this option is set, the kernel will send an NMI to all CPUs to dump
357their backtraces when a hung task is detected. This file shows up if
358CONFIG_DETECT_HUNG_TASK and CONFIG_SMP are enabled.
359
3600: Won't show all CPUs backtraces when a hung task is detected.
361This is the default behavior.
362
3631: Will non-maskably interrupt all CPUs and dump their backtraces when
364a hung task is detected.
365
366
367hung_task_panic
368===============
369
370Controls the kernel's behavior when a hung task is detected.
371This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
372
373= =================================================
3740 Continue operation. This is the default behavior.
3751 Panic immediately.
376= =================================================
377
378
379hung_task_check_count
380=====================
381
382The upper bound on the number of tasks that are checked.
383This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
384
385
386hung_task_timeout_secs
387======================
388
389When a task in D state did not get scheduled
390for more than this value report a warning.
391This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
392
3930 means infinite timeout, no checking is done.
394
395Possible values to set are in range {0:``LONG_MAX``/``HZ``}.
396
397
398hung_task_check_interval_secs
399=============================
400
401Hung task check interval. If hung task checking is enabled
402(see `hung_task_timeout_secs`_), the check is done every
403``hung_task_check_interval_secs`` seconds.
404This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
405
4060 (default) means use ``hung_task_timeout_secs`` as checking
407interval.
408
409Possible values to set are in range {0:``LONG_MAX``/``HZ``}.
410
411
412hung_task_warnings
413==================
414
415The maximum number of warnings to report. During a check interval
416if a hung task is detected, this value is decreased by 1.
417When this value reaches 0, no more warnings will be reported.
418This file shows up if ``CONFIG_DETECT_HUNG_TASK`` is enabled.
419
420-1: report an infinite number of warnings.
421
422
423hyperv_record_panic_msg
424=======================
425
426Controls whether the panic kmsg data should be reported to Hyper-V.
427
428= =========================================================
4290 Do not report panic kmsg data.
4301 Report the panic kmsg data. This is the default behavior.
431= =========================================================
432
433
434ignore-unaligned-usertrap
435=========================
436
437On architectures where unaligned accesses cause traps, and where this
438feature is supported (``CONFIG_SYSCTL_ARCH_UNALIGN_NO_WARN``;
439currently, ``arc``, ``ia64`` and ``loongarch``), controls whether all
440unaligned traps are logged.
441
442= =============================================================
4430 Log all unaligned accesses.
4441 Only warn the first time a process traps. This is the default
445 setting.
446= =============================================================
447
448See also `unaligned-trap`_ and `unaligned-dump-stack`_. On ``ia64``,
449this allows system administrators to override the
450``IA64_THREAD_UAC_NOPRINT`` ``prctl`` and avoid logs being flooded.
451
452
453kexec_load_disabled
454===================
455
456A toggle indicating if the ``kexec_load`` syscall has been disabled.
457This value defaults to 0 (false: ``kexec_load`` enabled), but can be
458set to 1 (true: ``kexec_load`` disabled).
459Once true, kexec can no longer be used, and the toggle cannot be set
460back to false.
461This allows a kexec image to be loaded before disabling the syscall,
462allowing a system to set up (and later use) an image without it being
463altered.
464Generally used together with the `modules_disabled`_ sysctl.
465
466
467kptr_restrict
468=============
469
470This toggle indicates whether restrictions are placed on
471exposing kernel addresses via ``/proc`` and other interfaces.
472
473When ``kptr_restrict`` is set to 0 (the default) the address is hashed
474before printing.
475(This is the equivalent to %p.)
476
477When ``kptr_restrict`` is set to 1, kernel pointers printed using the
478%pK format specifier will be replaced with 0s unless the user has
479``CAP_SYSLOG`` and effective user and group ids are equal to the real
480ids.
481This is because %pK checks are done at read() time rather than open()
482time, so if permissions are elevated between the open() and the read()
483(e.g via a setuid binary) then %pK will not leak kernel pointers to
484unprivileged users.
485Note, this is a temporary solution only.
486The correct long-term solution is to do the permission checks at
487open() time.
488Consider removing world read permissions from files that use %pK, and
489using `dmesg_restrict`_ to protect against uses of %pK in ``dmesg(8)``
490if leaking kernel pointer values to unprivileged users is a concern.
491
492When ``kptr_restrict`` is set to 2, kernel pointers printed using
493%pK will be replaced with 0s regardless of privileges.
494
495
496modprobe
497========
498
499The full path to the usermode helper for autoloading kernel modules,
500by default ``CONFIG_MODPROBE_PATH``, which in turn defaults to
501"/sbin/modprobe". This binary is executed when the kernel requests a
502module. For example, if userspace passes an unknown filesystem type
503to mount(), then the kernel will automatically request the
504corresponding filesystem module by executing this usermode helper.
505This usermode helper should insert the needed module into the kernel.
506
507This sysctl only affects module autoloading. It has no effect on the
508ability to explicitly insert modules.
509
510This sysctl can be used to debug module loading requests::
511
512 echo '#! /bin/sh' > /tmp/modprobe
513 echo 'echo "$@" >> /tmp/modprobe.log' >> /tmp/modprobe
514 echo 'exec /sbin/modprobe "$@"' >> /tmp/modprobe
515 chmod a+x /tmp/modprobe
516 echo /tmp/modprobe > /proc/sys/kernel/modprobe
517
518Alternatively, if this sysctl is set to the empty string, then module
519autoloading is completely disabled. The kernel will not try to
520execute a usermode helper at all, nor will it call the
521kernel_module_request LSM hook.
522
523If CONFIG_STATIC_USERMODEHELPER=y is set in the kernel configuration,
524then the configured static usermode helper overrides this sysctl,
525except that the empty string is still accepted to completely disable
526module autoloading as described above.
527
528modules_disabled
529================
530
531A toggle value indicating if modules are allowed to be loaded
532in an otherwise modular kernel. This toggle defaults to off
533(0), but can be set true (1). Once true, modules can be
534neither loaded nor unloaded, and the toggle cannot be set back
535to false. Generally used with the `kexec_load_disabled`_ toggle.
536
537
538.. _msgmni:
539
540msgmax, msgmnb, and msgmni
541==========================
542
543``msgmax`` is the maximum size of an IPC message, in bytes. 8192 by
544default (``MSGMAX``).
545
546``msgmnb`` is the maximum size of an IPC queue, in bytes. 16384 by
547default (``MSGMNB``).
548
549``msgmni`` is the maximum number of IPC queues. 32000 by default
550(``MSGMNI``).
551
552
553msg_next_id, sem_next_id, and shm_next_id (System V IPC)
554========================================================
555
556These three toggles allows to specify desired id for next allocated IPC
557object: message, semaphore or shared memory respectively.
558
559By default they are equal to -1, which means generic allocation logic.
560Possible values to set are in range {0:``INT_MAX``}.
561
562Notes:
563 1) kernel doesn't guarantee, that new object will have desired id. So,
564 it's up to userspace, how to handle an object with "wrong" id.
565 2) Toggle with non-default value will be set back to -1 by kernel after
566 successful IPC object allocation. If an IPC object allocation syscall
567 fails, it is undefined if the value remains unmodified or is reset to -1.
568
569
570ngroups_max
571===========
572
573Maximum number of supplementary groups, _i.e._ the maximum size which
574``setgroups`` will accept. Exports ``NGROUPS_MAX`` from the kernel.
575
576
577
578nmi_watchdog
579============
580
581This parameter can be used to control the NMI watchdog
582(i.e. the hard lockup detector) on x86 systems.
583
584= =================================
5850 Disable the hard lockup detector.
5861 Enable the hard lockup detector.
587= =================================
588
589The hard lockup detector monitors each CPU for its ability to respond to
590timer interrupts. The mechanism utilizes CPU performance counter registers
591that are programmed to generate Non-Maskable Interrupts (NMIs) periodically
592while a CPU is busy. Hence, the alternative name 'NMI watchdog'.
593
594The NMI watchdog is disabled by default if the kernel is running as a guest
595in a KVM virtual machine. This default can be overridden by adding::
596
597 nmi_watchdog=1
598
599to the guest kernel command line (see
600Documentation/admin-guide/kernel-parameters.rst).
601
602
603nmi_wd_lpm_factor (PPC only)
604============================
605
606Factor to apply to the NMI watchdog timeout (only when ``nmi_watchdog`` is
607set to 1). This factor represents the percentage added to
608``watchdog_thresh`` when calculating the NMI watchdog timeout during an
609LPM. The soft lockup timeout is not impacted.
610
611A value of 0 means no change. The default value is 200 meaning the NMI
612watchdog is set to 30s (based on ``watchdog_thresh`` equal to 10).
613
614
615numa_balancing
616==============
617
618Enables/disables and configures automatic page fault based NUMA memory
619balancing. Memory is moved automatically to nodes that access it often.
620The value to set can be the result of ORing the following:
621
622= =================================
6230 NUMA_BALANCING_DISABLED
6241 NUMA_BALANCING_NORMAL
6252 NUMA_BALANCING_MEMORY_TIERING
626= =================================
627
628Or NUMA_BALANCING_NORMAL to optimize page placement among different
629NUMA nodes to reduce remote accessing. On NUMA machines, there is a
630performance penalty if remote memory is accessed by a CPU. When this
631feature is enabled the kernel samples what task thread is accessing
632memory by periodically unmapping pages and later trapping a page
633fault. At the time of the page fault, it is determined if the data
634being accessed should be migrated to a local memory node.
635
636The unmapping of pages and trapping faults incur additional overhead that
637ideally is offset by improved memory locality but there is no universal
638guarantee. If the target workload is already bound to NUMA nodes then this
639feature should be disabled.
640
641Or NUMA_BALANCING_MEMORY_TIERING to optimize page placement among
642different types of memory (represented as different NUMA nodes) to
643place the hot pages in the fast memory. This is implemented based on
644unmapping and page fault too.
645
646numa_balancing_promote_rate_limit_MBps
647======================================
648
649Too high promotion/demotion throughput between different memory types
650may hurt application latency. This can be used to rate limit the
651promotion throughput. The per-node max promotion throughput in MB/s
652will be limited to be no more than the set value.
653
654A rule of thumb is to set this to less than 1/10 of the PMEM node
655write bandwidth.
656
657oops_all_cpu_backtrace
658======================
659
660If this option is set, the kernel will send an NMI to all CPUs to dump
661their backtraces when an oops event occurs. It should be used as a last
662resort in case a panic cannot be triggered (to protect VMs running, for
663example) or kdump can't be collected. This file shows up if CONFIG_SMP
664is enabled.
665
6660: Won't show all CPUs backtraces when an oops is detected.
667This is the default behavior.
668
6691: Will non-maskably interrupt all CPUs and dump their backtraces when
670an oops event is detected.
671
672
673oops_limit
674==========
675
676Number of kernel oopses after which the kernel should panic when
677``panic_on_oops`` is not set. Setting this to 0 disables checking
678the count. Setting this to 1 has the same effect as setting
679``panic_on_oops=1``. The default value is 10000.
680
681
682osrelease, ostype & version
683===========================
684
685::
686
687 # cat osrelease
688 2.1.88
689 # cat ostype
690 Linux
691 # cat version
692 #5 Wed Feb 25 21:49:24 MET 1998
693
694The files ``osrelease`` and ``ostype`` should be clear enough.
695``version``
696needs a little more clarification however. The '#5' means that
697this is the fifth kernel built from this source base and the
698date behind it indicates the time the kernel was built.
699The only way to tune these values is to rebuild the kernel :-)
700
701
702overflowgid & overflowuid
703=========================
704
705if your architecture did not always support 32-bit UIDs (i.e. arm,
706i386, m68k, sh, and sparc32), a fixed UID and GID will be returned to
707applications that use the old 16-bit UID/GID system calls, if the
708actual UID or GID would exceed 65535.
709
710These sysctls allow you to change the value of the fixed UID and GID.
711The default is 65534.
712
713
714panic
715=====
716
717The value in this file determines the behaviour of the kernel on a
718panic:
719
720* if zero, the kernel will loop forever;
721* if negative, the kernel will reboot immediately;
722* if positive, the kernel will reboot after the corresponding number
723 of seconds.
724
725When you use the software watchdog, the recommended setting is 60.
726
727
728panic_on_io_nmi
729===============
730
731Controls the kernel's behavior when a CPU receives an NMI caused by
732an IO error.
733
734= ==================================================================
7350 Try to continue operation (default).
7361 Panic immediately. The IO error triggered an NMI. This indicates a
737 serious system condition which could result in IO data corruption.
738 Rather than continuing, panicking might be a better choice. Some
739 servers issue this sort of NMI when the dump button is pushed,
740 and you can use this option to take a crash dump.
741= ==================================================================
742
743
744panic_on_oops
745=============
746
747Controls the kernel's behaviour when an oops or BUG is encountered.
748
749= ===================================================================
7500 Try to continue operation.
7511 Panic immediately. If the `panic` sysctl is also non-zero then the
752 machine will be rebooted.
753= ===================================================================
754
755
756panic_on_stackoverflow
757======================
758
759Controls the kernel's behavior when detecting the overflows of
760kernel, IRQ and exception stacks except a user stack.
761This file shows up if ``CONFIG_DEBUG_STACKOVERFLOW`` is enabled.
762
763= ==========================
7640 Try to continue operation.
7651 Panic immediately.
766= ==========================
767
768
769panic_on_unrecovered_nmi
770========================
771
772The default Linux behaviour on an NMI of either memory or unknown is
773to continue operation. For many environments such as scientific
774computing it is preferable that the box is taken out and the error
775dealt with than an uncorrected parity/ECC error get propagated.
776
777A small number of systems do generate NMIs for bizarre random reasons
778such as power management so the default is off. That sysctl works like
779the existing panic controls already in that directory.
780
781
782panic_on_warn
783=============
784
785Calls panic() in the WARN() path when set to 1. This is useful to avoid
786a kernel rebuild when attempting to kdump at the location of a WARN().
787
788= ================================================
7890 Only WARN(), default behaviour.
7901 Call panic() after printing out WARN() location.
791= ================================================
792
793
794panic_print
795===========
796
797Bitmask for printing system info when panic happens. User can chose
798combination of the following bits:
799
800===== ============================================
801bit 0 print all tasks info
802bit 1 print system memory info
803bit 2 print timer info
804bit 3 print locks info if ``CONFIG_LOCKDEP`` is on
805bit 4 print ftrace buffer
806bit 5 print all printk messages in buffer
807bit 6 print all CPUs backtrace (if available in the arch)
808===== ============================================
809
810So for example to print tasks and memory info on panic, user can::
811
812 echo 3 > /proc/sys/kernel/panic_print
813
814
815panic_on_rcu_stall
816==================
817
818When set to 1, calls panic() after RCU stall detection messages. This
819is useful to define the root cause of RCU stalls using a vmcore.
820
821= ============================================================
8220 Do not panic() when RCU stall takes place, default behavior.
8231 panic() after printing RCU stall messages.
824= ============================================================
825
826max_rcu_stall_to_panic
827======================
828
829When ``panic_on_rcu_stall`` is set to 1, this value determines the
830number of times that RCU can stall before panic() is called.
831
832When ``panic_on_rcu_stall`` is set to 0, this value is has no effect.
833
834perf_cpu_time_max_percent
835=========================
836
837Hints to the kernel how much CPU time it should be allowed to
838use to handle perf sampling events. If the perf subsystem
839is informed that its samples are exceeding this limit, it
840will drop its sampling frequency to attempt to reduce its CPU
841usage.
842
843Some perf sampling happens in NMIs. If these samples
844unexpectedly take too long to execute, the NMIs can become
845stacked up next to each other so much that nothing else is
846allowed to execute.
847
848===== ========================================================
8490 Disable the mechanism. Do not monitor or correct perf's
850 sampling rate no matter how CPU time it takes.
851
8521-100 Attempt to throttle perf's sample rate to this
853 percentage of CPU. Note: the kernel calculates an
854 "expected" length of each sample event. 100 here means
855 100% of that expected length. Even if this is set to
856 100, you may still see sample throttling if this
857 length is exceeded. Set to 0 if you truly do not care
858 how much CPU is consumed.
859===== ========================================================
860
861
862perf_event_paranoid
863===================
864
865Controls use of the performance events system by unprivileged
866users (without CAP_PERFMON). The default value is 2.
867
868For backward compatibility reasons access to system performance
869monitoring and observability remains open for CAP_SYS_ADMIN
870privileged processes but CAP_SYS_ADMIN usage for secure system
871performance monitoring and observability operations is discouraged
872with respect to CAP_PERFMON use cases.
873
874=== ==================================================================
875 -1 Allow use of (almost) all events by all users.
876
877 Ignore mlock limit after perf_event_mlock_kb without
878 ``CAP_IPC_LOCK``.
879
880>=0 Disallow ftrace function tracepoint by users without
881 ``CAP_PERFMON``.
882
883 Disallow raw tracepoint access by users without ``CAP_PERFMON``.
884
885>=1 Disallow CPU event access by users without ``CAP_PERFMON``.
886
887>=2 Disallow kernel profiling by users without ``CAP_PERFMON``.
888=== ==================================================================
889
890
891perf_event_max_stack
892====================
893
894Controls maximum number of stack frames to copy for (``attr.sample_type &
895PERF_SAMPLE_CALLCHAIN``) configured events, for instance, when using
896'``perf record -g``' or '``perf trace --call-graph fp``'.
897
898This can only be done when no events are in use that have callchains
899enabled, otherwise writing to this file will return ``-EBUSY``.
900
901The default value is 127.
902
903
904perf_event_mlock_kb
905===================
906
907Control size of per-cpu ring buffer not counted against mlock limit.
908
909The default value is 512 + 1 page
910
911
912perf_event_max_contexts_per_stack
913=================================
914
915Controls maximum number of stack frame context entries for
916(``attr.sample_type & PERF_SAMPLE_CALLCHAIN``) configured events, for
917instance, when using '``perf record -g``' or '``perf trace --call-graph fp``'.
918
919This can only be done when no events are in use that have callchains
920enabled, otherwise writing to this file will return ``-EBUSY``.
921
922The default value is 8.
923
924
925perf_user_access (arm64 only)
926=================================
927
928Controls user space access for reading perf event counters. When set to 1,
929user space can read performance monitor counter registers directly.
930
931The default value is 0 (access disabled).
932
933See Documentation/arm64/perf.rst for more information.
934
935
936pid_max
937=======
938
939PID allocation wrap value. When the kernel's next PID value
940reaches this value, it wraps back to a minimum PID value.
941PIDs of value ``pid_max`` or larger are not allocated.
942
943
944ns_last_pid
945===========
946
947The last pid allocated in the current (the one task using this sysctl
948lives in) pid namespace. When selecting a pid for a next task on fork
949kernel tries to allocate a number starting from this one.
950
951
952powersave-nap (PPC only)
953========================
954
955If set, Linux-PPC will use the 'nap' mode of powersaving,
956otherwise the 'doze' mode will be used.
957
958
959==============================================================
960
961printk
962======
963
964The four values in printk denote: ``console_loglevel``,
965``default_message_loglevel``, ``minimum_console_loglevel`` and
966``default_console_loglevel`` respectively.
967
968These values influence printk() behavior when printing or
969logging error messages. See '``man 2 syslog``' for more info on
970the different loglevels.
971
972======================== =====================================
973console_loglevel messages with a higher priority than
974 this will be printed to the console
975default_message_loglevel messages without an explicit priority
976 will be printed with this priority
977minimum_console_loglevel minimum (highest) value to which
978 console_loglevel can be set
979default_console_loglevel default value for console_loglevel
980======================== =====================================
981
982
983printk_delay
984============
985
986Delay each printk message in ``printk_delay`` milliseconds
987
988Value from 0 - 10000 is allowed.
989
990
991printk_ratelimit
992================
993
994Some warning messages are rate limited. ``printk_ratelimit`` specifies
995the minimum length of time between these messages (in seconds).
996The default value is 5 seconds.
997
998A value of 0 will disable rate limiting.
999
1000
1001printk_ratelimit_burst
1002======================
1003
1004While long term we enforce one message per `printk_ratelimit`_
1005seconds, we do allow a burst of messages to pass through.
1006``printk_ratelimit_burst`` specifies the number of messages we can
1007send before ratelimiting kicks in.
1008
1009The default value is 10 messages.
1010
1011
1012printk_devkmsg
1013==============
1014
1015Control the logging to ``/dev/kmsg`` from userspace:
1016
1017========= =============================================
1018ratelimit default, ratelimited
1019on unlimited logging to /dev/kmsg from userspace
1020off logging to /dev/kmsg disabled
1021========= =============================================
1022
1023The kernel command line parameter ``printk.devkmsg=`` overrides this and is
1024a one-time setting until next reboot: once set, it cannot be changed by
1025this sysctl interface anymore.
1026
1027==============================================================
1028
1029
1030pty
1031===
1032
1033See Documentation/filesystems/devpts.rst.
1034
1035
1036random
1037======
1038
1039This is a directory, with the following entries:
1040
1041* ``boot_id``: a UUID generated the first time this is retrieved, and
1042 unvarying after that;
1043
1044* ``uuid``: a UUID generated every time this is retrieved (this can
1045 thus be used to generate UUIDs at will);
1046
1047* ``entropy_avail``: the pool's entropy count, in bits;
1048
1049* ``poolsize``: the entropy pool size, in bits;
1050
1051* ``urandom_min_reseed_secs``: obsolete (used to determine the minimum
1052 number of seconds between urandom pool reseeding). This file is
1053 writable for compatibility purposes, but writing to it has no effect
1054 on any RNG behavior;
1055
1056* ``write_wakeup_threshold``: when the entropy count drops below this
1057 (as a number of bits), processes waiting to write to ``/dev/random``
1058 are woken up. This file is writable for compatibility purposes, but
1059 writing to it has no effect on any RNG behavior.
1060
1061
1062randomize_va_space
1063==================
1064
1065This option can be used to select the type of process address
1066space randomization that is used in the system, for architectures
1067that support this feature.
1068
1069== ===========================================================================
10700 Turn the process address space randomization off. This is the
1071 default for architectures that do not support this feature anyways,
1072 and kernels that are booted with the "norandmaps" parameter.
1073
10741 Make the addresses of mmap base, stack and VDSO page randomized.
1075 This, among other things, implies that shared libraries will be
1076 loaded to random addresses. Also for PIE-linked binaries, the
1077 location of code start is randomized. This is the default if the
1078 ``CONFIG_COMPAT_BRK`` option is enabled.
1079
10802 Additionally enable heap randomization. This is the default if
1081 ``CONFIG_COMPAT_BRK`` is disabled.
1082
1083 There are a few legacy applications out there (such as some ancient
1084 versions of libc.so.5 from 1996) that assume that brk area starts
1085 just after the end of the code+bss. These applications break when
1086 start of the brk area is randomized. There are however no known
1087 non-legacy applications that would be broken this way, so for most
1088 systems it is safe to choose full randomization.
1089
1090 Systems with ancient and/or broken binaries should be configured
1091 with ``CONFIG_COMPAT_BRK`` enabled, which excludes the heap from process
1092 address space randomization.
1093== ===========================================================================
1094
1095
1096real-root-dev
1097=============
1098
1099See Documentation/admin-guide/initrd.rst.
1100
1101
1102reboot-cmd (SPARC only)
1103=======================
1104
1105??? This seems to be a way to give an argument to the Sparc
1106ROM/Flash boot loader. Maybe to tell it what to do after
1107rebooting. ???
1108
1109
1110sched_energy_aware
1111==================
1112
1113Enables/disables Energy Aware Scheduling (EAS). EAS starts
1114automatically on platforms where it can run (that is,
1115platforms with asymmetric CPU topologies and having an Energy
1116Model available). If your platform happens to meet the
1117requirements for EAS but you do not want to use it, change
1118this value to 0.
1119
1120task_delayacct
1121===============
1122
1123Enables/disables task delay accounting (see
1124Documentation/accounting/delay-accounting.rst. Enabling this feature incurs
1125a small amount of overhead in the scheduler but is useful for debugging
1126and performance tuning. It is required by some tools such as iotop.
1127
1128sched_schedstats
1129================
1130
1131Enables/disables scheduler statistics. Enabling this feature
1132incurs a small amount of overhead in the scheduler but is
1133useful for debugging and performance tuning.
1134
1135sched_util_clamp_min
1136====================
1137
1138Max allowed *minimum* utilization.
1139
1140Default value is 1024, which is the maximum possible value.
1141
1142It means that any requested uclamp.min value cannot be greater than
1143sched_util_clamp_min, i.e., it is restricted to the range
1144[0:sched_util_clamp_min].
1145
1146sched_util_clamp_max
1147====================
1148
1149Max allowed *maximum* utilization.
1150
1151Default value is 1024, which is the maximum possible value.
1152
1153It means that any requested uclamp.max value cannot be greater than
1154sched_util_clamp_max, i.e., it is restricted to the range
1155[0:sched_util_clamp_max].
1156
1157sched_util_clamp_min_rt_default
1158===============================
1159
1160By default Linux is tuned for performance. Which means that RT tasks always run
1161at the highest frequency and most capable (highest capacity) CPU (in
1162heterogeneous systems).
1163
1164Uclamp achieves this by setting the requested uclamp.min of all RT tasks to
11651024 by default, which effectively boosts the tasks to run at the highest
1166frequency and biases them to run on the biggest CPU.
1167
1168This knob allows admins to change the default behavior when uclamp is being
1169used. In battery powered devices particularly, running at the maximum
1170capacity and frequency will increase energy consumption and shorten the battery
1171life.
1172
1173This knob is only effective for RT tasks which the user hasn't modified their
1174requested uclamp.min value via sched_setattr() syscall.
1175
1176This knob will not escape the range constraint imposed by sched_util_clamp_min
1177defined above.
1178
1179For example if
1180
1181 sched_util_clamp_min_rt_default = 800
1182 sched_util_clamp_min = 600
1183
1184Then the boost will be clamped to 600 because 800 is outside of the permissible
1185range of [0:600]. This could happen for instance if a powersave mode will
1186restrict all boosts temporarily by modifying sched_util_clamp_min. As soon as
1187this restriction is lifted, the requested sched_util_clamp_min_rt_default
1188will take effect.
1189
1190seccomp
1191=======
1192
1193See Documentation/userspace-api/seccomp_filter.rst.
1194
1195
1196sg-big-buff
1197===========
1198
1199This file shows the size of the generic SCSI (sg) buffer.
1200You can't tune it just yet, but you could change it on
1201compile time by editing ``include/scsi/sg.h`` and changing
1202the value of ``SG_BIG_BUFF``.
1203
1204There shouldn't be any reason to change this value. If
1205you can come up with one, you probably know what you
1206are doing anyway :)
1207
1208
1209shmall
1210======
1211
1212This parameter sets the total amount of shared memory pages that
1213can be used system wide. Hence, ``shmall`` should always be at least
1214``ceil(shmmax/PAGE_SIZE)``.
1215
1216If you are not sure what the default ``PAGE_SIZE`` is on your Linux
1217system, you can run the following command::
1218
1219 # getconf PAGE_SIZE
1220
1221
1222shmmax
1223======
1224
1225This value can be used to query and set the run time limit
1226on the maximum shared memory segment size that can be created.
1227Shared memory segments up to 1Gb are now supported in the
1228kernel. This value defaults to ``SHMMAX``.
1229
1230
1231shmmni
1232======
1233
1234This value determines the maximum number of shared memory segments.
12354096 by default (``SHMMNI``).
1236
1237
1238shm_rmid_forced
1239===============
1240
1241Linux lets you set resource limits, including how much memory one
1242process can consume, via ``setrlimit(2)``. Unfortunately, shared memory
1243segments are allowed to exist without association with any process, and
1244thus might not be counted against any resource limits. If enabled,
1245shared memory segments are automatically destroyed when their attach
1246count becomes zero after a detach or a process termination. It will
1247also destroy segments that were created, but never attached to, on exit
1248from the process. The only use left for ``IPC_RMID`` is to immediately
1249destroy an unattached segment. Of course, this breaks the way things are
1250defined, so some applications might stop working. Note that this
1251feature will do you no good unless you also configure your resource
1252limits (in particular, ``RLIMIT_AS`` and ``RLIMIT_NPROC``). Most systems don't
1253need this.
1254
1255Note that if you change this from 0 to 1, already created segments
1256without users and with a dead originative process will be destroyed.
1257
1258
1259sysctl_writes_strict
1260====================
1261
1262Control how file position affects the behavior of updating sysctl values
1263via the ``/proc/sys`` interface:
1264
1265 == ======================================================================
1266 -1 Legacy per-write sysctl value handling, with no printk warnings.
1267 Each write syscall must fully contain the sysctl value to be
1268 written, and multiple writes on the same sysctl file descriptor
1269 will rewrite the sysctl value, regardless of file position.
1270 0 Same behavior as above, but warn about processes that perform writes
1271 to a sysctl file descriptor when the file position is not 0.
1272 1 (default) Respect file position when writing sysctl strings. Multiple
1273 writes will append to the sysctl value buffer. Anything past the max
1274 length of the sysctl value buffer will be ignored. Writes to numeric
1275 sysctl entries must always be at file position 0 and the value must
1276 be fully contained in the buffer sent in the write syscall.
1277 == ======================================================================
1278
1279
1280softlockup_all_cpu_backtrace
1281============================
1282
1283This value controls the soft lockup detector thread's behavior
1284when a soft lockup condition is detected as to whether or not
1285to gather further debug information. If enabled, each cpu will
1286be issued an NMI and instructed to capture stack trace.
1287
1288This feature is only applicable for architectures which support
1289NMI.
1290
1291= ============================================
12920 Do nothing. This is the default behavior.
12931 On detection capture more debug information.
1294= ============================================
1295
1296
1297softlockup_panic
1298=================
1299
1300This parameter can be used to control whether the kernel panics
1301when a soft lockup is detected.
1302
1303= ============================================
13040 Don't panic on soft lockup.
13051 Panic on soft lockup.
1306= ============================================
1307
1308This can also be set using the softlockup_panic kernel parameter.
1309
1310
1311soft_watchdog
1312=============
1313
1314This parameter can be used to control the soft lockup detector.
1315
1316= =================================
13170 Disable the soft lockup detector.
13181 Enable the soft lockup detector.
1319= =================================
1320
1321The soft lockup detector monitors CPUs for threads that are hogging the CPUs
1322without rescheduling voluntarily, and thus prevent the 'migration/N' threads
1323from running, causing the watchdog work fail to execute. The mechanism depends
1324on the CPUs ability to respond to timer interrupts which are needed for the
1325watchdog work to be queued by the watchdog timer function, otherwise the NMI
1326watchdog — if enabled — can detect a hard lockup condition.
1327
1328
1329split_lock_mitigate (x86 only)
1330==============================
1331
1332On x86, each "split lock" imposes a system-wide performance penalty. On larger
1333systems, large numbers of split locks from unprivileged users can result in
1334denials of service to well-behaved and potentially more important users.
1335
1336The kernel mitigates these bad users by detecting split locks and imposing
1337penalties: forcing them to wait and only allowing one core to execute split
1338locks at a time.
1339
1340These mitigations can make those bad applications unbearably slow. Setting
1341split_lock_mitigate=0 may restore some application performance, but will also
1342increase system exposure to denial of service attacks from split lock users.
1343
1344= ===================================================================
13450 Disable the mitigation mode - just warns the split lock on kernel log
1346 and exposes the system to denials of service from the split lockers.
13471 Enable the mitigation mode (this is the default) - penalizes the split
1348 lockers with intentional performance degradation.
1349= ===================================================================
1350
1351
1352stack_erasing
1353=============
1354
1355This parameter can be used to control kernel stack erasing at the end
1356of syscalls for kernels built with ``CONFIG_GCC_PLUGIN_STACKLEAK``.
1357
1358That erasing reduces the information which kernel stack leak bugs
1359can reveal and blocks some uninitialized stack variable attacks.
1360The tradeoff is the performance impact: on a single CPU system kernel
1361compilation sees a 1% slowdown, other systems and workloads may vary.
1362
1363= ====================================================================
13640 Kernel stack erasing is disabled, STACKLEAK_METRICS are not updated.
13651 Kernel stack erasing is enabled (default), it is performed before
1366 returning to the userspace at the end of syscalls.
1367= ====================================================================
1368
1369
1370stop-a (SPARC only)
1371===================
1372
1373Controls Stop-A:
1374
1375= ====================================
13760 Stop-A has no effect.
13771 Stop-A breaks to the PROM (default).
1378= ====================================
1379
1380Stop-A is always enabled on a panic, so that the user can return to
1381the boot PROM.
1382
1383
1384sysrq
1385=====
1386
1387See Documentation/admin-guide/sysrq.rst.
1388
1389
1390tainted
1391=======
1392
1393Non-zero if the kernel has been tainted. Numeric values, which can be
1394ORed together. The letters are seen in "Tainted" line of Oops reports.
1395
1396====== ===== ==============================================================
1397 1 `(P)` proprietary module was loaded
1398 2 `(F)` module was force loaded
1399 4 `(S)` kernel running on an out of specification system
1400 8 `(R)` module was force unloaded
1401 16 `(M)` processor reported a Machine Check Exception (MCE)
1402 32 `(B)` bad page referenced or some unexpected page flags
1403 64 `(U)` taint requested by userspace application
1404 128 `(D)` kernel died recently, i.e. there was an OOPS or BUG
1405 256 `(A)` an ACPI table was overridden by user
1406 512 `(W)` kernel issued warning
1407 1024 `(C)` staging driver was loaded
1408 2048 `(I)` workaround for bug in platform firmware applied
1409 4096 `(O)` externally-built ("out-of-tree") module was loaded
1410 8192 `(E)` unsigned module was loaded
1411 16384 `(L)` soft lockup occurred
1412 32768 `(K)` kernel has been live patched
1413 65536 `(X)` Auxiliary taint, defined and used by for distros
1414131072 `(T)` The kernel was built with the struct randomization plugin
1415====== ===== ==============================================================
1416
1417See Documentation/admin-guide/tainted-kernels.rst for more information.
1418
1419Note:
1420 writes to this sysctl interface will fail with ``EINVAL`` if the kernel is
1421 booted with the command line option ``panic_on_taint=<bitmask>,nousertaint``
1422 and any of the ORed together values being written to ``tainted`` match with
1423 the bitmask declared on panic_on_taint.
1424 See Documentation/admin-guide/kernel-parameters.rst for more details on
1425 that particular kernel command line option and its optional
1426 ``nousertaint`` switch.
1427
1428threads-max
1429===========
1430
1431This value controls the maximum number of threads that can be created
1432using ``fork()``.
1433
1434During initialization the kernel sets this value such that even if the
1435maximum number of threads is created, the thread structures occupy only
1436a part (1/8th) of the available RAM pages.
1437
1438The minimum value that can be written to ``threads-max`` is 1.
1439
1440The maximum value that can be written to ``threads-max`` is given by the
1441constant ``FUTEX_TID_MASK`` (0x3fffffff).
1442
1443If a value outside of this range is written to ``threads-max`` an
1444``EINVAL`` error occurs.
1445
1446
1447traceoff_on_warning
1448===================
1449
1450When set, disables tracing (see Documentation/trace/ftrace.rst) when a
1451``WARN()`` is hit.
1452
1453
1454tracepoint_printk
1455=================
1456
1457When tracepoints are sent to printk() (enabled by the ``tp_printk``
1458boot parameter), this entry provides runtime control::
1459
1460 echo 0 > /proc/sys/kernel/tracepoint_printk
1461
1462will stop tracepoints from being sent to printk(), and::
1463
1464 echo 1 > /proc/sys/kernel/tracepoint_printk
1465
1466will send them to printk() again.
1467
1468This only works if the kernel was booted with ``tp_printk`` enabled.
1469
1470See Documentation/admin-guide/kernel-parameters.rst and
1471Documentation/trace/boottime-trace.rst.
1472
1473
1474.. _unaligned-dump-stack:
1475
1476unaligned-dump-stack (ia64)
1477===========================
1478
1479When logging unaligned accesses, controls whether the stack is
1480dumped.
1481
1482= ===================================================
14830 Do not dump the stack. This is the default setting.
14841 Dump the stack.
1485= ===================================================
1486
1487See also `ignore-unaligned-usertrap`_.
1488
1489
1490unaligned-trap
1491==============
1492
1493On architectures where unaligned accesses cause traps, and where this
1494feature is supported (``CONFIG_SYSCTL_ARCH_UNALIGN_ALLOW``; currently,
1495``arc``, ``parisc`` and ``loongarch``), controls whether unaligned traps
1496are caught and emulated (instead of failing).
1497
1498= ========================================================
14990 Do not emulate unaligned accesses.
15001 Emulate unaligned accesses. This is the default setting.
1501= ========================================================
1502
1503See also `ignore-unaligned-usertrap`_.
1504
1505
1506unknown_nmi_panic
1507=================
1508
1509The value in this file affects behavior of handling NMI. When the
1510value is non-zero, unknown NMI is trapped and then panic occurs. At
1511that time, kernel debugging information is displayed on console.
1512
1513NMI switch that most IA32 servers have fires unknown NMI up, for
1514example. If a system hangs up, try pressing the NMI switch.
1515
1516
1517unprivileged_bpf_disabled
1518=========================
1519
1520Writing 1 to this entry will disable unprivileged calls to ``bpf()``;
1521once disabled, calling ``bpf()`` without ``CAP_SYS_ADMIN`` or ``CAP_BPF``
1522will return ``-EPERM``. Once set to 1, this can't be cleared from the
1523running kernel anymore.
1524
1525Writing 2 to this entry will also disable unprivileged calls to ``bpf()``,
1526however, an admin can still change this setting later on, if needed, by
1527writing 0 or 1 to this entry.
1528
1529If ``BPF_UNPRIV_DEFAULT_OFF`` is enabled in the kernel config, then this
1530entry will default to 2 instead of 0.
1531
1532= =============================================================
15330 Unprivileged calls to ``bpf()`` are enabled
15341 Unprivileged calls to ``bpf()`` are disabled without recovery
15352 Unprivileged calls to ``bpf()`` are disabled
1536= =============================================================
1537
1538
1539warn_limit
1540==========
1541
1542Number of kernel warnings after which the kernel should panic when
1543``panic_on_warn`` is not set. Setting this to 0 disables checking
1544the warning count. Setting this to 1 has the same effect as setting
1545``panic_on_warn=1``. The default value is 0.
1546
1547
1548watchdog
1549========
1550
1551This parameter can be used to disable or enable the soft lockup detector
1552*and* the NMI watchdog (i.e. the hard lockup detector) at the same time.
1553
1554= ==============================
15550 Disable both lockup detectors.
15561 Enable both lockup detectors.
1557= ==============================
1558
1559The soft lockup detector and the NMI watchdog can also be disabled or
1560enabled individually, using the ``soft_watchdog`` and ``nmi_watchdog``
1561parameters.
1562If the ``watchdog`` parameter is read, for example by executing::
1563
1564 cat /proc/sys/kernel/watchdog
1565
1566the output of this command (0 or 1) shows the logical OR of
1567``soft_watchdog`` and ``nmi_watchdog``.
1568
1569
1570watchdog_cpumask
1571================
1572
1573This value can be used to control on which cpus the watchdog may run.
1574The default cpumask is all possible cores, but if ``NO_HZ_FULL`` is
1575enabled in the kernel config, and cores are specified with the
1576``nohz_full=`` boot argument, those cores are excluded by default.
1577Offline cores can be included in this mask, and if the core is later
1578brought online, the watchdog will be started based on the mask value.
1579
1580Typically this value would only be touched in the ``nohz_full`` case
1581to re-enable cores that by default were not running the watchdog,
1582if a kernel lockup was suspected on those cores.
1583
1584The argument value is the standard cpulist format for cpumasks,
1585so for example to enable the watchdog on cores 0, 2, 3, and 4 you
1586might say::
1587
1588 echo 0,2-4 > /proc/sys/kernel/watchdog_cpumask
1589
1590
1591watchdog_thresh
1592===============
1593
1594This value can be used to control the frequency of hrtimer and NMI
1595events and the soft and hard lockup thresholds. The default threshold
1596is 10 seconds.
1597
1598The softlockup threshold is (``2 * watchdog_thresh``). Setting this
1599tunable to zero will disable lockup detection altogether.