Loading...
1Documentation for /proc/sys/kernel/* kernel version 2.2.10
2 (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
3 (c) 2009, Shen Feng<shen@cn.fujitsu.com>
4
5For general info and legal blurb, please look in README.
6
7==============================================================
8
9This file contains documentation for the sysctl files in
10/proc/sys/kernel/ and is valid for Linux kernel version 2.2.
11
12The files in this directory can be used to tune and monitor
13miscellaneous and general things in the operation of the Linux
14kernel. Since some of the files _can_ be used to screw up your
15system, it is advisable to read both documentation and source
16before actually making adjustments.
17
18Currently, these files might (depending on your configuration)
19show up in /proc/sys/kernel:
20
21- acct
22- acpi_video_flags
23- auto_msgmni
24- bootloader_type [ X86 only ]
25- bootloader_version [ X86 only ]
26- callhome [ S390 only ]
27- cap_last_cap
28- core_pattern
29- core_pipe_limit
30- core_uses_pid
31- ctrl-alt-del
32- dmesg_restrict
33- domainname
34- hostname
35- hotplug
36- hardlockup_all_cpu_backtrace
37- hung_task_panic
38- hung_task_check_count
39- hung_task_timeout_secs
40- hung_task_warnings
41- kexec_load_disabled
42- kptr_restrict
43- kstack_depth_to_print [ X86 only ]
44- l2cr [ PPC only ]
45- modprobe ==> Documentation/debugging-modules.txt
46- modules_disabled
47- msg_next_id [ sysv ipc ]
48- msgmax
49- msgmnb
50- msgmni
51- nmi_watchdog
52- osrelease
53- ostype
54- overflowgid
55- overflowuid
56- panic
57- panic_on_oops
58- panic_on_stackoverflow
59- panic_on_unrecovered_nmi
60- panic_on_warn
61- perf_cpu_time_max_percent
62- perf_event_paranoid
63- pid_max
64- powersave-nap [ PPC only ]
65- printk
66- printk_delay
67- printk_ratelimit
68- printk_ratelimit_burst
69- pty ==> Documentation/filesystems/devpts.txt
70- randomize_va_space
71- real-root-dev ==> Documentation/initrd.txt
72- reboot-cmd [ SPARC only ]
73- rtsig-max
74- rtsig-nr
75- sem
76- sem_next_id [ sysv ipc ]
77- sg-big-buff [ generic SCSI device (sg) ]
78- shm_next_id [ sysv ipc ]
79- shm_rmid_forced
80- shmall
81- shmmax [ sysv ipc ]
82- shmmni
83- softlockup_all_cpu_backtrace
84- soft_watchdog
85- stop-a [ SPARC only ]
86- sysrq ==> Documentation/sysrq.txt
87- sysctl_writes_strict
88- tainted
89- threads-max
90- unknown_nmi_panic
91- watchdog
92- watchdog_thresh
93- version
94
95==============================================================
96
97acct:
98
99highwater lowwater frequency
100
101If BSD-style process accounting is enabled these values control
102its behaviour. If free space on filesystem where the log lives
103goes below <lowwater>% accounting suspends. If free space gets
104above <highwater>% accounting resumes. <Frequency> determines
105how often do we check the amount of free space (value is in
106seconds). Default:
1074 2 30
108That is, suspend accounting if there left <= 2% free; resume it
109if we got >=4%; consider information about amount of free space
110valid for 30 seconds.
111
112==============================================================
113
114acpi_video_flags:
115
116flags
117
118See Doc*/kernel/power/video.txt, it allows mode of video boot to be
119set during run time.
120
121==============================================================
122
123auto_msgmni:
124
125This variable has no effect and may be removed in future kernel
126releases. Reading it always returns 0.
127Up to Linux 3.17, it enabled/disabled automatic recomputing of msgmni
128upon memory add/remove or upon ipc namespace creation/removal.
129Echoing "1" into this file enabled msgmni automatic recomputing.
130Echoing "0" turned it off. auto_msgmni default value was 1.
131
132
133==============================================================
134
135bootloader_type:
136
137x86 bootloader identification
138
139This gives the bootloader type number as indicated by the bootloader,
140shifted left by 4, and OR'd with the low four bits of the bootloader
141version. The reason for this encoding is that this used to match the
142type_of_loader field in the kernel header; the encoding is kept for
143backwards compatibility. That is, if the full bootloader type number
144is 0x15 and the full version number is 0x234, this file will contain
145the value 340 = 0x154.
146
147See the type_of_loader and ext_loader_type fields in
148Documentation/x86/boot.txt for additional information.
149
150==============================================================
151
152bootloader_version:
153
154x86 bootloader version
155
156The complete bootloader version number. In the example above, this
157file will contain the value 564 = 0x234.
158
159See the type_of_loader and ext_loader_ver fields in
160Documentation/x86/boot.txt for additional information.
161
162==============================================================
163
164callhome:
165
166Controls the kernel's callhome behavior in case of a kernel panic.
167
168The s390 hardware allows an operating system to send a notification
169to a service organization (callhome) in case of an operating system panic.
170
171When the value in this file is 0 (which is the default behavior)
172nothing happens in case of a kernel panic. If this value is set to "1"
173the complete kernel oops message is send to the IBM customer service
174organization in case the mainframe the Linux operating system is running
175on has a service contract with IBM.
176
177==============================================================
178
179cap_last_cap
180
181Highest valid capability of the running kernel. Exports
182CAP_LAST_CAP from the kernel.
183
184==============================================================
185
186core_pattern:
187
188core_pattern is used to specify a core dumpfile pattern name.
189. max length 128 characters; default value is "core"
190. core_pattern is used as a pattern template for the output filename;
191 certain string patterns (beginning with '%') are substituted with
192 their actual values.
193. backward compatibility with core_uses_pid:
194 If core_pattern does not include "%p" (default does not)
195 and core_uses_pid is set, then .PID will be appended to
196 the filename.
197. corename format specifiers:
198 %<NUL> '%' is dropped
199 %% output one '%'
200 %p pid
201 %P global pid (init PID namespace)
202 %i tid
203 %I global tid (init PID namespace)
204 %u uid (in initial user namespace)
205 %g gid (in initial user namespace)
206 %d dump mode, matches PR_SET_DUMPABLE and
207 /proc/sys/fs/suid_dumpable
208 %s signal number
209 %t UNIX time of dump
210 %h hostname
211 %e executable filename (may be shortened)
212 %E executable path
213 %<OTHER> both are dropped
214. If the first character of the pattern is a '|', the kernel will treat
215 the rest of the pattern as a command to run. The core dump will be
216 written to the standard input of that program instead of to a file.
217
218==============================================================
219
220core_pipe_limit:
221
222This sysctl is only applicable when core_pattern is configured to pipe
223core files to a user space helper (when the first character of
224core_pattern is a '|', see above). When collecting cores via a pipe
225to an application, it is occasionally useful for the collecting
226application to gather data about the crashing process from its
227/proc/pid directory. In order to do this safely, the kernel must wait
228for the collecting process to exit, so as not to remove the crashing
229processes proc files prematurely. This in turn creates the
230possibility that a misbehaving userspace collecting process can block
231the reaping of a crashed process simply by never exiting. This sysctl
232defends against that. It defines how many concurrent crashing
233processes may be piped to user space applications in parallel. If
234this value is exceeded, then those crashing processes above that value
235are noted via the kernel log and their cores are skipped. 0 is a
236special value, indicating that unlimited processes may be captured in
237parallel, but that no waiting will take place (i.e. the collecting
238process is not guaranteed access to /proc/<crashing pid>/). This
239value defaults to 0.
240
241==============================================================
242
243core_uses_pid:
244
245The default coredump filename is "core". By setting
246core_uses_pid to 1, the coredump filename becomes core.PID.
247If core_pattern does not include "%p" (default does not)
248and core_uses_pid is set, then .PID will be appended to
249the filename.
250
251==============================================================
252
253ctrl-alt-del:
254
255When the value in this file is 0, ctrl-alt-del is trapped and
256sent to the init(1) program to handle a graceful restart.
257When, however, the value is > 0, Linux's reaction to a Vulcan
258Nerve Pinch (tm) will be an immediate reboot, without even
259syncing its dirty buffers.
260
261Note: when a program (like dosemu) has the keyboard in 'raw'
262mode, the ctrl-alt-del is intercepted by the program before it
263ever reaches the kernel tty layer, and it's up to the program
264to decide what to do with it.
265
266==============================================================
267
268dmesg_restrict:
269
270This toggle indicates whether unprivileged users are prevented
271from using dmesg(8) to view messages from the kernel's log buffer.
272When dmesg_restrict is set to (0) there are no restrictions. When
273dmesg_restrict is set set to (1), users must have CAP_SYSLOG to use
274dmesg(8).
275
276The kernel config option CONFIG_SECURITY_DMESG_RESTRICT sets the
277default value of dmesg_restrict.
278
279==============================================================
280
281domainname & hostname:
282
283These files can be used to set the NIS/YP domainname and the
284hostname of your box in exactly the same way as the commands
285domainname and hostname, i.e.:
286# echo "darkstar" > /proc/sys/kernel/hostname
287# echo "mydomain" > /proc/sys/kernel/domainname
288has the same effect as
289# hostname "darkstar"
290# domainname "mydomain"
291
292Note, however, that the classic darkstar.frop.org has the
293hostname "darkstar" and DNS (Internet Domain Name Server)
294domainname "frop.org", not to be confused with the NIS (Network
295Information Service) or YP (Yellow Pages) domainname. These two
296domain names are in general different. For a detailed discussion
297see the hostname(1) man page.
298
299==============================================================
300hardlockup_all_cpu_backtrace:
301
302This value controls the hard lockup detector behavior when a hard
303lockup condition is detected as to whether or not to gather further
304debug information. If enabled, arch-specific all-CPU stack dumping
305will be initiated.
306
3070: do nothing. This is the default behavior.
308
3091: on detection capture more debug information.
310==============================================================
311
312hotplug:
313
314Path for the hotplug policy agent.
315Default value is "/sbin/hotplug".
316
317==============================================================
318
319hung_task_panic:
320
321Controls the kernel's behavior when a hung task is detected.
322This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
323
3240: continue operation. This is the default behavior.
325
3261: panic immediately.
327
328==============================================================
329
330hung_task_check_count:
331
332The upper bound on the number of tasks that are checked.
333This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
334
335==============================================================
336
337hung_task_timeout_secs:
338
339Check interval. When a task in D state did not get scheduled
340for more than this value report a warning.
341This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
342
3430: means infinite timeout - no checking done.
344Possible values to set are in range {0..LONG_MAX/HZ}.
345
346==============================================================
347
348hung_task_warnings:
349
350The maximum number of warnings to report. During a check interval
351if a hung task is detected, this value is decreased by 1.
352When this value reaches 0, no more warnings will be reported.
353This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
354
355-1: report an infinite number of warnings.
356
357==============================================================
358
359kexec_load_disabled:
360
361A toggle indicating if the kexec_load syscall has been disabled. This
362value defaults to 0 (false: kexec_load enabled), but can be set to 1
363(true: kexec_load disabled). Once true, kexec can no longer be used, and
364the toggle cannot be set back to false. This allows a kexec image to be
365loaded before disabling the syscall, allowing a system to set up (and
366later use) an image without it being altered. Generally used together
367with the "modules_disabled" sysctl.
368
369==============================================================
370
371kptr_restrict:
372
373This toggle indicates whether restrictions are placed on
374exposing kernel addresses via /proc and other interfaces.
375
376When kptr_restrict is set to (0), the default, there are no restrictions.
377
378When kptr_restrict is set to (1), kernel pointers printed using the %pK
379format specifier will be replaced with 0's unless the user has CAP_SYSLOG
380and effective user and group ids are equal to the real ids. This is
381because %pK checks are done at read() time rather than open() time, so
382if permissions are elevated between the open() and the read() (e.g via
383a setuid binary) then %pK will not leak kernel pointers to unprivileged
384users. Note, this is a temporary solution only. The correct long-term
385solution is to do the permission checks at open() time. Consider removing
386world read permissions from files that use %pK, and using dmesg_restrict
387to protect against uses of %pK in dmesg(8) if leaking kernel pointer
388values to unprivileged users is a concern.
389
390When kptr_restrict is set to (2), kernel pointers printed using
391%pK will be replaced with 0's regardless of privileges.
392
393==============================================================
394
395kstack_depth_to_print: (X86 only)
396
397Controls the number of words to print when dumping the raw
398kernel stack.
399
400==============================================================
401
402l2cr: (PPC only)
403
404This flag controls the L2 cache of G3 processor boards. If
4050, the cache is disabled. Enabled if nonzero.
406
407==============================================================
408
409modules_disabled:
410
411A toggle value indicating if modules are allowed to be loaded
412in an otherwise modular kernel. This toggle defaults to off
413(0), but can be set true (1). Once true, modules can be
414neither loaded nor unloaded, and the toggle cannot be set back
415to false. Generally used with the "kexec_load_disabled" toggle.
416
417==============================================================
418
419msg_next_id, sem_next_id, and shm_next_id:
420
421These three toggles allows to specify desired id for next allocated IPC
422object: message, semaphore or shared memory respectively.
423
424By default they are equal to -1, which means generic allocation logic.
425Possible values to set are in range {0..INT_MAX}.
426
427Notes:
4281) kernel doesn't guarantee, that new object will have desired id. So,
429it's up to userspace, how to handle an object with "wrong" id.
4302) Toggle with non-default value will be set back to -1 by kernel after
431successful IPC object allocation.
432
433==============================================================
434
435nmi_watchdog:
436
437This parameter can be used to control the NMI watchdog
438(i.e. the hard lockup detector) on x86 systems.
439
440 0 - disable the hard lockup detector
441 1 - enable the hard lockup detector
442
443The hard lockup detector monitors each CPU for its ability to respond to
444timer interrupts. The mechanism utilizes CPU performance counter registers
445that are programmed to generate Non-Maskable Interrupts (NMIs) periodically
446while a CPU is busy. Hence, the alternative name 'NMI watchdog'.
447
448The NMI watchdog is disabled by default if the kernel is running as a guest
449in a KVM virtual machine. This default can be overridden by adding
450
451 nmi_watchdog=1
452
453to the guest kernel command line (see Documentation/kernel-parameters.txt).
454
455==============================================================
456
457numa_balancing
458
459Enables/disables automatic page fault based NUMA memory
460balancing. Memory is moved automatically to nodes
461that access it often.
462
463Enables/disables automatic NUMA memory balancing. On NUMA machines, there
464is a performance penalty if remote memory is accessed by a CPU. When this
465feature is enabled the kernel samples what task thread is accessing memory
466by periodically unmapping pages and later trapping a page fault. At the
467time of the page fault, it is determined if the data being accessed should
468be migrated to a local memory node.
469
470The unmapping of pages and trapping faults incur additional overhead that
471ideally is offset by improved memory locality but there is no universal
472guarantee. If the target workload is already bound to NUMA nodes then this
473feature should be disabled. Otherwise, if the system overhead from the
474feature is too high then the rate the kernel samples for NUMA hinting
475faults may be controlled by the numa_balancing_scan_period_min_ms,
476numa_balancing_scan_delay_ms, numa_balancing_scan_period_max_ms,
477numa_balancing_scan_size_mb, and numa_balancing_settle_count sysctls.
478
479==============================================================
480
481numa_balancing_scan_period_min_ms, numa_balancing_scan_delay_ms,
482numa_balancing_scan_period_max_ms, numa_balancing_scan_size_mb
483
484Automatic NUMA balancing scans tasks address space and unmaps pages to
485detect if pages are properly placed or if the data should be migrated to a
486memory node local to where the task is running. Every "scan delay" the task
487scans the next "scan size" number of pages in its address space. When the
488end of the address space is reached the scanner restarts from the beginning.
489
490In combination, the "scan delay" and "scan size" determine the scan rate.
491When "scan delay" decreases, the scan rate increases. The scan delay and
492hence the scan rate of every task is adaptive and depends on historical
493behaviour. If pages are properly placed then the scan delay increases,
494otherwise the scan delay decreases. The "scan size" is not adaptive but
495the higher the "scan size", the higher the scan rate.
496
497Higher scan rates incur higher system overhead as page faults must be
498trapped and potentially data must be migrated. However, the higher the scan
499rate, the more quickly a tasks memory is migrated to a local node if the
500workload pattern changes and minimises performance impact due to remote
501memory accesses. These sysctls control the thresholds for scan delays and
502the number of pages scanned.
503
504numa_balancing_scan_period_min_ms is the minimum time in milliseconds to
505scan a tasks virtual memory. It effectively controls the maximum scanning
506rate for each task.
507
508numa_balancing_scan_delay_ms is the starting "scan delay" used for a task
509when it initially forks.
510
511numa_balancing_scan_period_max_ms is the maximum time in milliseconds to
512scan a tasks virtual memory. It effectively controls the minimum scanning
513rate for each task.
514
515numa_balancing_scan_size_mb is how many megabytes worth of pages are
516scanned for a given scan.
517
518==============================================================
519
520osrelease, ostype & version:
521
522# cat osrelease
5232.1.88
524# cat ostype
525Linux
526# cat version
527#5 Wed Feb 25 21:49:24 MET 1998
528
529The files osrelease and ostype should be clear enough. Version
530needs a little more clarification however. The '#5' means that
531this is the fifth kernel built from this source base and the
532date behind it indicates the time the kernel was built.
533The only way to tune these values is to rebuild the kernel :-)
534
535==============================================================
536
537overflowgid & overflowuid:
538
539if your architecture did not always support 32-bit UIDs (i.e. arm,
540i386, m68k, sh, and sparc32), a fixed UID and GID will be returned to
541applications that use the old 16-bit UID/GID system calls, if the
542actual UID or GID would exceed 65535.
543
544These sysctls allow you to change the value of the fixed UID and GID.
545The default is 65534.
546
547==============================================================
548
549panic:
550
551The value in this file represents the number of seconds the kernel
552waits before rebooting on a panic. When you use the software watchdog,
553the recommended setting is 60.
554
555==============================================================
556
557panic_on_io_nmi:
558
559Controls the kernel's behavior when a CPU receives an NMI caused by
560an IO error.
561
5620: try to continue operation (default)
563
5641: panic immediately. The IO error triggered an NMI. This indicates a
565 serious system condition which could result in IO data corruption.
566 Rather than continuing, panicking might be a better choice. Some
567 servers issue this sort of NMI when the dump button is pushed,
568 and you can use this option to take a crash dump.
569
570==============================================================
571
572panic_on_oops:
573
574Controls the kernel's behaviour when an oops or BUG is encountered.
575
5760: try to continue operation
577
5781: panic immediately. If the `panic' sysctl is also non-zero then the
579 machine will be rebooted.
580
581==============================================================
582
583panic_on_stackoverflow:
584
585Controls the kernel's behavior when detecting the overflows of
586kernel, IRQ and exception stacks except a user stack.
587This file shows up if CONFIG_DEBUG_STACKOVERFLOW is enabled.
588
5890: try to continue operation.
590
5911: panic immediately.
592
593==============================================================
594
595panic_on_unrecovered_nmi:
596
597The default Linux behaviour on an NMI of either memory or unknown is
598to continue operation. For many environments such as scientific
599computing it is preferable that the box is taken out and the error
600dealt with than an uncorrected parity/ECC error get propagated.
601
602A small number of systems do generate NMI's for bizarre random reasons
603such as power management so the default is off. That sysctl works like
604the existing panic controls already in that directory.
605
606==============================================================
607
608panic_on_warn:
609
610Calls panic() in the WARN() path when set to 1. This is useful to avoid
611a kernel rebuild when attempting to kdump at the location of a WARN().
612
6130: only WARN(), default behaviour.
614
6151: call panic() after printing out WARN() location.
616
617==============================================================
618
619perf_cpu_time_max_percent:
620
621Hints to the kernel how much CPU time it should be allowed to
622use to handle perf sampling events. If the perf subsystem
623is informed that its samples are exceeding this limit, it
624will drop its sampling frequency to attempt to reduce its CPU
625usage.
626
627Some perf sampling happens in NMIs. If these samples
628unexpectedly take too long to execute, the NMIs can become
629stacked up next to each other so much that nothing else is
630allowed to execute.
631
6320: disable the mechanism. Do not monitor or correct perf's
633 sampling rate no matter how CPU time it takes.
634
6351-100: attempt to throttle perf's sample rate to this
636 percentage of CPU. Note: the kernel calculates an
637 "expected" length of each sample event. 100 here means
638 100% of that expected length. Even if this is set to
639 100, you may still see sample throttling if this
640 length is exceeded. Set to 0 if you truly do not care
641 how much CPU is consumed.
642
643==============================================================
644
645perf_event_paranoid:
646
647Controls use of the performance events system by unprivileged
648users (without CAP_SYS_ADMIN). The default value is 2.
649
650 -1: Allow use of (almost) all events by all users
651>=0: Disallow raw tracepoint access by users without CAP_IOC_LOCK
652>=1: Disallow CPU event access by users without CAP_SYS_ADMIN
653>=2: Disallow kernel profiling by users without CAP_SYS_ADMIN
654
655==============================================================
656
657pid_max:
658
659PID allocation wrap value. When the kernel's next PID value
660reaches this value, it wraps back to a minimum PID value.
661PIDs of value pid_max or larger are not allocated.
662
663==============================================================
664
665ns_last_pid:
666
667The last pid allocated in the current (the one task using this sysctl
668lives in) pid namespace. When selecting a pid for a next task on fork
669kernel tries to allocate a number starting from this one.
670
671==============================================================
672
673powersave-nap: (PPC only)
674
675If set, Linux-PPC will use the 'nap' mode of powersaving,
676otherwise the 'doze' mode will be used.
677
678==============================================================
679
680printk:
681
682The four values in printk denote: console_loglevel,
683default_message_loglevel, minimum_console_loglevel and
684default_console_loglevel respectively.
685
686These values influence printk() behavior when printing or
687logging error messages. See 'man 2 syslog' for more info on
688the different loglevels.
689
690- console_loglevel: messages with a higher priority than
691 this will be printed to the console
692- default_message_loglevel: messages without an explicit priority
693 will be printed with this priority
694- minimum_console_loglevel: minimum (highest) value to which
695 console_loglevel can be set
696- default_console_loglevel: default value for console_loglevel
697
698==============================================================
699
700printk_delay:
701
702Delay each printk message in printk_delay milliseconds
703
704Value from 0 - 10000 is allowed.
705
706==============================================================
707
708printk_ratelimit:
709
710Some warning messages are rate limited. printk_ratelimit specifies
711the minimum length of time between these messages (in jiffies), by
712default we allow one every 5 seconds.
713
714A value of 0 will disable rate limiting.
715
716==============================================================
717
718printk_ratelimit_burst:
719
720While long term we enforce one message per printk_ratelimit
721seconds, we do allow a burst of messages to pass through.
722printk_ratelimit_burst specifies the number of messages we can
723send before ratelimiting kicks in.
724
725==============================================================
726
727randomize_va_space:
728
729This option can be used to select the type of process address
730space randomization that is used in the system, for architectures
731that support this feature.
732
7330 - Turn the process address space randomization off. This is the
734 default for architectures that do not support this feature anyways,
735 and kernels that are booted with the "norandmaps" parameter.
736
7371 - Make the addresses of mmap base, stack and VDSO page randomized.
738 This, among other things, implies that shared libraries will be
739 loaded to random addresses. Also for PIE-linked binaries, the
740 location of code start is randomized. This is the default if the
741 CONFIG_COMPAT_BRK option is enabled.
742
7432 - Additionally enable heap randomization. This is the default if
744 CONFIG_COMPAT_BRK is disabled.
745
746 There are a few legacy applications out there (such as some ancient
747 versions of libc.so.5 from 1996) that assume that brk area starts
748 just after the end of the code+bss. These applications break when
749 start of the brk area is randomized. There are however no known
750 non-legacy applications that would be broken this way, so for most
751 systems it is safe to choose full randomization.
752
753 Systems with ancient and/or broken binaries should be configured
754 with CONFIG_COMPAT_BRK enabled, which excludes the heap from process
755 address space randomization.
756
757==============================================================
758
759reboot-cmd: (Sparc only)
760
761??? This seems to be a way to give an argument to the Sparc
762ROM/Flash boot loader. Maybe to tell it what to do after
763rebooting. ???
764
765==============================================================
766
767rtsig-max & rtsig-nr:
768
769The file rtsig-max can be used to tune the maximum number
770of POSIX realtime (queued) signals that can be outstanding
771in the system.
772
773rtsig-nr shows the number of RT signals currently queued.
774
775==============================================================
776
777sched_schedstats:
778
779Enables/disables scheduler statistics. Enabling this feature
780incurs a small amount of overhead in the scheduler but is
781useful for debugging and performance tuning.
782
783==============================================================
784
785sg-big-buff:
786
787This file shows the size of the generic SCSI (sg) buffer.
788You can't tune it just yet, but you could change it on
789compile time by editing include/scsi/sg.h and changing
790the value of SG_BIG_BUFF.
791
792There shouldn't be any reason to change this value. If
793you can come up with one, you probably know what you
794are doing anyway :)
795
796==============================================================
797
798shmall:
799
800This parameter sets the total amount of shared memory pages that
801can be used system wide. Hence, SHMALL should always be at least
802ceil(shmmax/PAGE_SIZE).
803
804If you are not sure what the default PAGE_SIZE is on your Linux
805system, you can run the following command:
806
807# getconf PAGE_SIZE
808
809==============================================================
810
811shmmax:
812
813This value can be used to query and set the run time limit
814on the maximum shared memory segment size that can be created.
815Shared memory segments up to 1Gb are now supported in the
816kernel. This value defaults to SHMMAX.
817
818==============================================================
819
820shm_rmid_forced:
821
822Linux lets you set resource limits, including how much memory one
823process can consume, via setrlimit(2). Unfortunately, shared memory
824segments are allowed to exist without association with any process, and
825thus might not be counted against any resource limits. If enabled,
826shared memory segments are automatically destroyed when their attach
827count becomes zero after a detach or a process termination. It will
828also destroy segments that were created, but never attached to, on exit
829from the process. The only use left for IPC_RMID is to immediately
830destroy an unattached segment. Of course, this breaks the way things are
831defined, so some applications might stop working. Note that this
832feature will do you no good unless you also configure your resource
833limits (in particular, RLIMIT_AS and RLIMIT_NPROC). Most systems don't
834need this.
835
836Note that if you change this from 0 to 1, already created segments
837without users and with a dead originative process will be destroyed.
838
839==============================================================
840
841sysctl_writes_strict:
842
843Control how file position affects the behavior of updating sysctl values
844via the /proc/sys interface:
845
846 -1 - Legacy per-write sysctl value handling, with no printk warnings.
847 Each write syscall must fully contain the sysctl value to be
848 written, and multiple writes on the same sysctl file descriptor
849 will rewrite the sysctl value, regardless of file position.
850 0 - Same behavior as above, but warn about processes that perform writes
851 to a sysctl file descriptor when the file position is not 0.
852 1 - (default) Respect file position when writing sysctl strings. Multiple
853 writes will append to the sysctl value buffer. Anything past the max
854 length of the sysctl value buffer will be ignored. Writes to numeric
855 sysctl entries must always be at file position 0 and the value must
856 be fully contained in the buffer sent in the write syscall.
857
858==============================================================
859
860softlockup_all_cpu_backtrace:
861
862This value controls the soft lockup detector thread's behavior
863when a soft lockup condition is detected as to whether or not
864to gather further debug information. If enabled, each cpu will
865be issued an NMI and instructed to capture stack trace.
866
867This feature is only applicable for architectures which support
868NMI.
869
8700: do nothing. This is the default behavior.
871
8721: on detection capture more debug information.
873
874==============================================================
875
876soft_watchdog
877
878This parameter can be used to control the soft lockup detector.
879
880 0 - disable the soft lockup detector
881 1 - enable the soft lockup detector
882
883The soft lockup detector monitors CPUs for threads that are hogging the CPUs
884without rescheduling voluntarily, and thus prevent the 'watchdog/N' threads
885from running. The mechanism depends on the CPUs ability to respond to timer
886interrupts which are needed for the 'watchdog/N' threads to be woken up by
887the watchdog timer function, otherwise the NMI watchdog - if enabled - can
888detect a hard lockup condition.
889
890==============================================================
891
892tainted:
893
894Non-zero if the kernel has been tainted. Numeric values, which
895can be ORed together:
896
897 1 - A module with a non-GPL license has been loaded, this
898 includes modules with no license.
899 Set by modutils >= 2.4.9 and module-init-tools.
900 2 - A module was force loaded by insmod -f.
901 Set by modutils >= 2.4.9 and module-init-tools.
902 4 - Unsafe SMP processors: SMP with CPUs not designed for SMP.
903 8 - A module was forcibly unloaded from the system by rmmod -f.
904 16 - A hardware machine check error occurred on the system.
905 32 - A bad page was discovered on the system.
906 64 - The user has asked that the system be marked "tainted". This
907 could be because they are running software that directly modifies
908 the hardware, or for other reasons.
909 128 - The system has died.
910 256 - The ACPI DSDT has been overridden with one supplied by the user
911 instead of using the one provided by the hardware.
912 512 - A kernel warning has occurred.
9131024 - A module from drivers/staging was loaded.
9142048 - The system is working around a severe firmware bug.
9154096 - An out-of-tree module has been loaded.
9168192 - An unsigned module has been loaded in a kernel supporting module
917 signature.
91816384 - A soft lockup has previously occurred on the system.
91932768 - The kernel has been live patched.
920
921==============================================================
922
923threads-max
924
925This value controls the maximum number of threads that can be created
926using fork().
927
928During initialization the kernel sets this value such that even if the
929maximum number of threads is created, the thread structures occupy only
930a part (1/8th) of the available RAM pages.
931
932The minimum value that can be written to threads-max is 20.
933The maximum value that can be written to threads-max is given by the
934constant FUTEX_TID_MASK (0x3fffffff).
935If a value outside of this range is written to threads-max an error
936EINVAL occurs.
937
938The value written is checked against the available RAM pages. If the
939thread structures would occupy too much (more than 1/8th) of the
940available RAM pages threads-max is reduced accordingly.
941
942==============================================================
943
944unknown_nmi_panic:
945
946The value in this file affects behavior of handling NMI. When the
947value is non-zero, unknown NMI is trapped and then panic occurs. At
948that time, kernel debugging information is displayed on console.
949
950NMI switch that most IA32 servers have fires unknown NMI up, for
951example. If a system hangs up, try pressing the NMI switch.
952
953==============================================================
954
955watchdog:
956
957This parameter can be used to disable or enable the soft lockup detector
958_and_ the NMI watchdog (i.e. the hard lockup detector) at the same time.
959
960 0 - disable both lockup detectors
961 1 - enable both lockup detectors
962
963The soft lockup detector and the NMI watchdog can also be disabled or
964enabled individually, using the soft_watchdog and nmi_watchdog parameters.
965If the watchdog parameter is read, for example by executing
966
967 cat /proc/sys/kernel/watchdog
968
969the output of this command (0 or 1) shows the logical OR of soft_watchdog
970and nmi_watchdog.
971
972==============================================================
973
974watchdog_cpumask:
975
976This value can be used to control on which cpus the watchdog may run.
977The default cpumask is all possible cores, but if NO_HZ_FULL is
978enabled in the kernel config, and cores are specified with the
979nohz_full= boot argument, those cores are excluded by default.
980Offline cores can be included in this mask, and if the core is later
981brought online, the watchdog will be started based on the mask value.
982
983Typically this value would only be touched in the nohz_full case
984to re-enable cores that by default were not running the watchdog,
985if a kernel lockup was suspected on those cores.
986
987The argument value is the standard cpulist format for cpumasks,
988so for example to enable the watchdog on cores 0, 2, 3, and 4 you
989might say:
990
991 echo 0,2-4 > /proc/sys/kernel/watchdog_cpumask
992
993==============================================================
994
995watchdog_thresh:
996
997This value can be used to control the frequency of hrtimer and NMI
998events and the soft and hard lockup thresholds. The default threshold
999is 10 seconds.
1000
1001The softlockup threshold is (2 * watchdog_thresh). Setting this
1002tunable to zero will disable lockup detection altogether.
1003
1004==============================================================
1Documentation for /proc/sys/kernel/* kernel version 2.2.10
2 (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
3 (c) 2009, Shen Feng<shen@cn.fujitsu.com>
4
5For general info and legal blurb, please look in README.
6
7==============================================================
8
9This file contains documentation for the sysctl files in
10/proc/sys/kernel/ and is valid for Linux kernel version 2.2.
11
12The files in this directory can be used to tune and monitor
13miscellaneous and general things in the operation of the Linux
14kernel. Since some of the files _can_ be used to screw up your
15system, it is advisable to read both documentation and source
16before actually making adjustments.
17
18Currently, these files might (depending on your configuration)
19show up in /proc/sys/kernel:
20
21- acct
22- acpi_video_flags
23- auto_msgmni
24- bootloader_type [ X86 only ]
25- bootloader_version [ X86 only ]
26- callhome [ S390 only ]
27- cap_last_cap
28- core_pattern
29- core_pipe_limit
30- core_uses_pid
31- ctrl-alt-del
32- dmesg_restrict
33- domainname
34- hostname
35- hotplug
36- hardlockup_all_cpu_backtrace
37- hung_task_panic
38- hung_task_check_count
39- hung_task_timeout_secs
40- hung_task_warnings
41- kexec_load_disabled
42- kptr_restrict
43- l2cr [ PPC only ]
44- modprobe ==> Documentation/debugging-modules.txt
45- modules_disabled
46- msg_next_id [ sysv ipc ]
47- msgmax
48- msgmnb
49- msgmni
50- nmi_watchdog
51- osrelease
52- ostype
53- overflowgid
54- overflowuid
55- panic
56- panic_on_oops
57- panic_on_stackoverflow
58- panic_on_unrecovered_nmi
59- panic_on_warn
60- panic_on_rcu_stall
61- perf_cpu_time_max_percent
62- perf_event_paranoid
63- perf_event_max_stack
64- perf_event_max_contexts_per_stack
65- pid_max
66- powersave-nap [ PPC only ]
67- printk
68- printk_delay
69- printk_ratelimit
70- printk_ratelimit_burst
71- pty ==> Documentation/filesystems/devpts.txt
72- randomize_va_space
73- real-root-dev ==> Documentation/admin-guide/initrd.rst
74- reboot-cmd [ SPARC only ]
75- rtsig-max
76- rtsig-nr
77- sem
78- sem_next_id [ sysv ipc ]
79- sg-big-buff [ generic SCSI device (sg) ]
80- shm_next_id [ sysv ipc ]
81- shm_rmid_forced
82- shmall
83- shmmax [ sysv ipc ]
84- shmmni
85- softlockup_all_cpu_backtrace
86- soft_watchdog
87- stop-a [ SPARC only ]
88- sysrq ==> Documentation/sysrq.txt
89- sysctl_writes_strict
90- tainted
91- threads-max
92- unknown_nmi_panic
93- watchdog
94- watchdog_thresh
95- version
96
97==============================================================
98
99acct:
100
101highwater lowwater frequency
102
103If BSD-style process accounting is enabled these values control
104its behaviour. If free space on filesystem where the log lives
105goes below <lowwater>% accounting suspends. If free space gets
106above <highwater>% accounting resumes. <Frequency> determines
107how often do we check the amount of free space (value is in
108seconds). Default:
1094 2 30
110That is, suspend accounting if there left <= 2% free; resume it
111if we got >=4%; consider information about amount of free space
112valid for 30 seconds.
113
114==============================================================
115
116acpi_video_flags:
117
118flags
119
120See Doc*/kernel/power/video.txt, it allows mode of video boot to be
121set during run time.
122
123==============================================================
124
125auto_msgmni:
126
127This variable has no effect and may be removed in future kernel
128releases. Reading it always returns 0.
129Up to Linux 3.17, it enabled/disabled automatic recomputing of msgmni
130upon memory add/remove or upon ipc namespace creation/removal.
131Echoing "1" into this file enabled msgmni automatic recomputing.
132Echoing "0" turned it off. auto_msgmni default value was 1.
133
134
135==============================================================
136
137bootloader_type:
138
139x86 bootloader identification
140
141This gives the bootloader type number as indicated by the bootloader,
142shifted left by 4, and OR'd with the low four bits of the bootloader
143version. The reason for this encoding is that this used to match the
144type_of_loader field in the kernel header; the encoding is kept for
145backwards compatibility. That is, if the full bootloader type number
146is 0x15 and the full version number is 0x234, this file will contain
147the value 340 = 0x154.
148
149See the type_of_loader and ext_loader_type fields in
150Documentation/x86/boot.txt for additional information.
151
152==============================================================
153
154bootloader_version:
155
156x86 bootloader version
157
158The complete bootloader version number. In the example above, this
159file will contain the value 564 = 0x234.
160
161See the type_of_loader and ext_loader_ver fields in
162Documentation/x86/boot.txt for additional information.
163
164==============================================================
165
166callhome:
167
168Controls the kernel's callhome behavior in case of a kernel panic.
169
170The s390 hardware allows an operating system to send a notification
171to a service organization (callhome) in case of an operating system panic.
172
173When the value in this file is 0 (which is the default behavior)
174nothing happens in case of a kernel panic. If this value is set to "1"
175the complete kernel oops message is send to the IBM customer service
176organization in case the mainframe the Linux operating system is running
177on has a service contract with IBM.
178
179==============================================================
180
181cap_last_cap
182
183Highest valid capability of the running kernel. Exports
184CAP_LAST_CAP from the kernel.
185
186==============================================================
187
188core_pattern:
189
190core_pattern is used to specify a core dumpfile pattern name.
191. max length 128 characters; default value is "core"
192. core_pattern is used as a pattern template for the output filename;
193 certain string patterns (beginning with '%') are substituted with
194 their actual values.
195. backward compatibility with core_uses_pid:
196 If core_pattern does not include "%p" (default does not)
197 and core_uses_pid is set, then .PID will be appended to
198 the filename.
199. corename format specifiers:
200 %<NUL> '%' is dropped
201 %% output one '%'
202 %p pid
203 %P global pid (init PID namespace)
204 %i tid
205 %I global tid (init PID namespace)
206 %u uid (in initial user namespace)
207 %g gid (in initial user namespace)
208 %d dump mode, matches PR_SET_DUMPABLE and
209 /proc/sys/fs/suid_dumpable
210 %s signal number
211 %t UNIX time of dump
212 %h hostname
213 %e executable filename (may be shortened)
214 %E executable path
215 %<OTHER> both are dropped
216. If the first character of the pattern is a '|', the kernel will treat
217 the rest of the pattern as a command to run. The core dump will be
218 written to the standard input of that program instead of to a file.
219
220==============================================================
221
222core_pipe_limit:
223
224This sysctl is only applicable when core_pattern is configured to pipe
225core files to a user space helper (when the first character of
226core_pattern is a '|', see above). When collecting cores via a pipe
227to an application, it is occasionally useful for the collecting
228application to gather data about the crashing process from its
229/proc/pid directory. In order to do this safely, the kernel must wait
230for the collecting process to exit, so as not to remove the crashing
231processes proc files prematurely. This in turn creates the
232possibility that a misbehaving userspace collecting process can block
233the reaping of a crashed process simply by never exiting. This sysctl
234defends against that. It defines how many concurrent crashing
235processes may be piped to user space applications in parallel. If
236this value is exceeded, then those crashing processes above that value
237are noted via the kernel log and their cores are skipped. 0 is a
238special value, indicating that unlimited processes may be captured in
239parallel, but that no waiting will take place (i.e. the collecting
240process is not guaranteed access to /proc/<crashing pid>/). This
241value defaults to 0.
242
243==============================================================
244
245core_uses_pid:
246
247The default coredump filename is "core". By setting
248core_uses_pid to 1, the coredump filename becomes core.PID.
249If core_pattern does not include "%p" (default does not)
250and core_uses_pid is set, then .PID will be appended to
251the filename.
252
253==============================================================
254
255ctrl-alt-del:
256
257When the value in this file is 0, ctrl-alt-del is trapped and
258sent to the init(1) program to handle a graceful restart.
259When, however, the value is > 0, Linux's reaction to a Vulcan
260Nerve Pinch (tm) will be an immediate reboot, without even
261syncing its dirty buffers.
262
263Note: when a program (like dosemu) has the keyboard in 'raw'
264mode, the ctrl-alt-del is intercepted by the program before it
265ever reaches the kernel tty layer, and it's up to the program
266to decide what to do with it.
267
268==============================================================
269
270dmesg_restrict:
271
272This toggle indicates whether unprivileged users are prevented
273from using dmesg(8) to view messages from the kernel's log buffer.
274When dmesg_restrict is set to (0) there are no restrictions. When
275dmesg_restrict is set set to (1), users must have CAP_SYSLOG to use
276dmesg(8).
277
278The kernel config option CONFIG_SECURITY_DMESG_RESTRICT sets the
279default value of dmesg_restrict.
280
281==============================================================
282
283domainname & hostname:
284
285These files can be used to set the NIS/YP domainname and the
286hostname of your box in exactly the same way as the commands
287domainname and hostname, i.e.:
288# echo "darkstar" > /proc/sys/kernel/hostname
289# echo "mydomain" > /proc/sys/kernel/domainname
290has the same effect as
291# hostname "darkstar"
292# domainname "mydomain"
293
294Note, however, that the classic darkstar.frop.org has the
295hostname "darkstar" and DNS (Internet Domain Name Server)
296domainname "frop.org", not to be confused with the NIS (Network
297Information Service) or YP (Yellow Pages) domainname. These two
298domain names are in general different. For a detailed discussion
299see the hostname(1) man page.
300
301==============================================================
302hardlockup_all_cpu_backtrace:
303
304This value controls the hard lockup detector behavior when a hard
305lockup condition is detected as to whether or not to gather further
306debug information. If enabled, arch-specific all-CPU stack dumping
307will be initiated.
308
3090: do nothing. This is the default behavior.
310
3111: on detection capture more debug information.
312==============================================================
313
314hotplug:
315
316Path for the hotplug policy agent.
317Default value is "/sbin/hotplug".
318
319==============================================================
320
321hung_task_panic:
322
323Controls the kernel's behavior when a hung task is detected.
324This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
325
3260: continue operation. This is the default behavior.
327
3281: panic immediately.
329
330==============================================================
331
332hung_task_check_count:
333
334The upper bound on the number of tasks that are checked.
335This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
336
337==============================================================
338
339hung_task_timeout_secs:
340
341Check interval. When a task in D state did not get scheduled
342for more than this value report a warning.
343This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
344
3450: means infinite timeout - no checking done.
346Possible values to set are in range {0..LONG_MAX/HZ}.
347
348==============================================================
349
350hung_task_warnings:
351
352The maximum number of warnings to report. During a check interval
353if a hung task is detected, this value is decreased by 1.
354When this value reaches 0, no more warnings will be reported.
355This file shows up if CONFIG_DETECT_HUNG_TASK is enabled.
356
357-1: report an infinite number of warnings.
358
359==============================================================
360
361kexec_load_disabled:
362
363A toggle indicating if the kexec_load syscall has been disabled. This
364value defaults to 0 (false: kexec_load enabled), but can be set to 1
365(true: kexec_load disabled). Once true, kexec can no longer be used, and
366the toggle cannot be set back to false. This allows a kexec image to be
367loaded before disabling the syscall, allowing a system to set up (and
368later use) an image without it being altered. Generally used together
369with the "modules_disabled" sysctl.
370
371==============================================================
372
373kptr_restrict:
374
375This toggle indicates whether restrictions are placed on
376exposing kernel addresses via /proc and other interfaces.
377
378When kptr_restrict is set to (0), the default, there are no restrictions.
379
380When kptr_restrict is set to (1), kernel pointers printed using the %pK
381format specifier will be replaced with 0's unless the user has CAP_SYSLOG
382and effective user and group ids are equal to the real ids. This is
383because %pK checks are done at read() time rather than open() time, so
384if permissions are elevated between the open() and the read() (e.g via
385a setuid binary) then %pK will not leak kernel pointers to unprivileged
386users. Note, this is a temporary solution only. The correct long-term
387solution is to do the permission checks at open() time. Consider removing
388world read permissions from files that use %pK, and using dmesg_restrict
389to protect against uses of %pK in dmesg(8) if leaking kernel pointer
390values to unprivileged users is a concern.
391
392When kptr_restrict is set to (2), kernel pointers printed using
393%pK will be replaced with 0's regardless of privileges.
394
395==============================================================
396
397l2cr: (PPC only)
398
399This flag controls the L2 cache of G3 processor boards. If
4000, the cache is disabled. Enabled if nonzero.
401
402==============================================================
403
404modules_disabled:
405
406A toggle value indicating if modules are allowed to be loaded
407in an otherwise modular kernel. This toggle defaults to off
408(0), but can be set true (1). Once true, modules can be
409neither loaded nor unloaded, and the toggle cannot be set back
410to false. Generally used with the "kexec_load_disabled" toggle.
411
412==============================================================
413
414msg_next_id, sem_next_id, and shm_next_id:
415
416These three toggles allows to specify desired id for next allocated IPC
417object: message, semaphore or shared memory respectively.
418
419By default they are equal to -1, which means generic allocation logic.
420Possible values to set are in range {0..INT_MAX}.
421
422Notes:
4231) kernel doesn't guarantee, that new object will have desired id. So,
424it's up to userspace, how to handle an object with "wrong" id.
4252) Toggle with non-default value will be set back to -1 by kernel after
426successful IPC object allocation.
427
428==============================================================
429
430nmi_watchdog:
431
432This parameter can be used to control the NMI watchdog
433(i.e. the hard lockup detector) on x86 systems.
434
435 0 - disable the hard lockup detector
436 1 - enable the hard lockup detector
437
438The hard lockup detector monitors each CPU for its ability to respond to
439timer interrupts. The mechanism utilizes CPU performance counter registers
440that are programmed to generate Non-Maskable Interrupts (NMIs) periodically
441while a CPU is busy. Hence, the alternative name 'NMI watchdog'.
442
443The NMI watchdog is disabled by default if the kernel is running as a guest
444in a KVM virtual machine. This default can be overridden by adding
445
446 nmi_watchdog=1
447
448to the guest kernel command line (see Documentation/admin-guide/kernel-parameters.rst).
449
450==============================================================
451
452numa_balancing
453
454Enables/disables automatic page fault based NUMA memory
455balancing. Memory is moved automatically to nodes
456that access it often.
457
458Enables/disables automatic NUMA memory balancing. On NUMA machines, there
459is a performance penalty if remote memory is accessed by a CPU. When this
460feature is enabled the kernel samples what task thread is accessing memory
461by periodically unmapping pages and later trapping a page fault. At the
462time of the page fault, it is determined if the data being accessed should
463be migrated to a local memory node.
464
465The unmapping of pages and trapping faults incur additional overhead that
466ideally is offset by improved memory locality but there is no universal
467guarantee. If the target workload is already bound to NUMA nodes then this
468feature should be disabled. Otherwise, if the system overhead from the
469feature is too high then the rate the kernel samples for NUMA hinting
470faults may be controlled by the numa_balancing_scan_period_min_ms,
471numa_balancing_scan_delay_ms, numa_balancing_scan_period_max_ms,
472numa_balancing_scan_size_mb, and numa_balancing_settle_count sysctls.
473
474==============================================================
475
476numa_balancing_scan_period_min_ms, numa_balancing_scan_delay_ms,
477numa_balancing_scan_period_max_ms, numa_balancing_scan_size_mb
478
479Automatic NUMA balancing scans tasks address space and unmaps pages to
480detect if pages are properly placed or if the data should be migrated to a
481memory node local to where the task is running. Every "scan delay" the task
482scans the next "scan size" number of pages in its address space. When the
483end of the address space is reached the scanner restarts from the beginning.
484
485In combination, the "scan delay" and "scan size" determine the scan rate.
486When "scan delay" decreases, the scan rate increases. The scan delay and
487hence the scan rate of every task is adaptive and depends on historical
488behaviour. If pages are properly placed then the scan delay increases,
489otherwise the scan delay decreases. The "scan size" is not adaptive but
490the higher the "scan size", the higher the scan rate.
491
492Higher scan rates incur higher system overhead as page faults must be
493trapped and potentially data must be migrated. However, the higher the scan
494rate, the more quickly a tasks memory is migrated to a local node if the
495workload pattern changes and minimises performance impact due to remote
496memory accesses. These sysctls control the thresholds for scan delays and
497the number of pages scanned.
498
499numa_balancing_scan_period_min_ms is the minimum time in milliseconds to
500scan a tasks virtual memory. It effectively controls the maximum scanning
501rate for each task.
502
503numa_balancing_scan_delay_ms is the starting "scan delay" used for a task
504when it initially forks.
505
506numa_balancing_scan_period_max_ms is the maximum time in milliseconds to
507scan a tasks virtual memory. It effectively controls the minimum scanning
508rate for each task.
509
510numa_balancing_scan_size_mb is how many megabytes worth of pages are
511scanned for a given scan.
512
513==============================================================
514
515osrelease, ostype & version:
516
517# cat osrelease
5182.1.88
519# cat ostype
520Linux
521# cat version
522#5 Wed Feb 25 21:49:24 MET 1998
523
524The files osrelease and ostype should be clear enough. Version
525needs a little more clarification however. The '#5' means that
526this is the fifth kernel built from this source base and the
527date behind it indicates the time the kernel was built.
528The only way to tune these values is to rebuild the kernel :-)
529
530==============================================================
531
532overflowgid & overflowuid:
533
534if your architecture did not always support 32-bit UIDs (i.e. arm,
535i386, m68k, sh, and sparc32), a fixed UID and GID will be returned to
536applications that use the old 16-bit UID/GID system calls, if the
537actual UID or GID would exceed 65535.
538
539These sysctls allow you to change the value of the fixed UID and GID.
540The default is 65534.
541
542==============================================================
543
544panic:
545
546The value in this file represents the number of seconds the kernel
547waits before rebooting on a panic. When you use the software watchdog,
548the recommended setting is 60.
549
550==============================================================
551
552panic_on_io_nmi:
553
554Controls the kernel's behavior when a CPU receives an NMI caused by
555an IO error.
556
5570: try to continue operation (default)
558
5591: panic immediately. The IO error triggered an NMI. This indicates a
560 serious system condition which could result in IO data corruption.
561 Rather than continuing, panicking might be a better choice. Some
562 servers issue this sort of NMI when the dump button is pushed,
563 and you can use this option to take a crash dump.
564
565==============================================================
566
567panic_on_oops:
568
569Controls the kernel's behaviour when an oops or BUG is encountered.
570
5710: try to continue operation
572
5731: panic immediately. If the `panic' sysctl is also non-zero then the
574 machine will be rebooted.
575
576==============================================================
577
578panic_on_stackoverflow:
579
580Controls the kernel's behavior when detecting the overflows of
581kernel, IRQ and exception stacks except a user stack.
582This file shows up if CONFIG_DEBUG_STACKOVERFLOW is enabled.
583
5840: try to continue operation.
585
5861: panic immediately.
587
588==============================================================
589
590panic_on_unrecovered_nmi:
591
592The default Linux behaviour on an NMI of either memory or unknown is
593to continue operation. For many environments such as scientific
594computing it is preferable that the box is taken out and the error
595dealt with than an uncorrected parity/ECC error get propagated.
596
597A small number of systems do generate NMI's for bizarre random reasons
598such as power management so the default is off. That sysctl works like
599the existing panic controls already in that directory.
600
601==============================================================
602
603panic_on_warn:
604
605Calls panic() in the WARN() path when set to 1. This is useful to avoid
606a kernel rebuild when attempting to kdump at the location of a WARN().
607
6080: only WARN(), default behaviour.
609
6101: call panic() after printing out WARN() location.
611
612==============================================================
613
614panic_on_rcu_stall:
615
616When set to 1, calls panic() after RCU stall detection messages. This
617is useful to define the root cause of RCU stalls using a vmcore.
618
6190: do not panic() when RCU stall takes place, default behavior.
620
6211: panic() after printing RCU stall messages.
622
623==============================================================
624
625perf_cpu_time_max_percent:
626
627Hints to the kernel how much CPU time it should be allowed to
628use to handle perf sampling events. If the perf subsystem
629is informed that its samples are exceeding this limit, it
630will drop its sampling frequency to attempt to reduce its CPU
631usage.
632
633Some perf sampling happens in NMIs. If these samples
634unexpectedly take too long to execute, the NMIs can become
635stacked up next to each other so much that nothing else is
636allowed to execute.
637
6380: disable the mechanism. Do not monitor or correct perf's
639 sampling rate no matter how CPU time it takes.
640
6411-100: attempt to throttle perf's sample rate to this
642 percentage of CPU. Note: the kernel calculates an
643 "expected" length of each sample event. 100 here means
644 100% of that expected length. Even if this is set to
645 100, you may still see sample throttling if this
646 length is exceeded. Set to 0 if you truly do not care
647 how much CPU is consumed.
648
649==============================================================
650
651perf_event_paranoid:
652
653Controls use of the performance events system by unprivileged
654users (without CAP_SYS_ADMIN). The default value is 2.
655
656 -1: Allow use of (almost) all events by all users
657>=0: Disallow raw tracepoint access by users without CAP_IOC_LOCK
658>=1: Disallow CPU event access by users without CAP_SYS_ADMIN
659>=2: Disallow kernel profiling by users without CAP_SYS_ADMIN
660
661==============================================================
662
663perf_event_max_stack:
664
665Controls maximum number of stack frames to copy for (attr.sample_type &
666PERF_SAMPLE_CALLCHAIN) configured events, for instance, when using
667'perf record -g' or 'perf trace --call-graph fp'.
668
669This can only be done when no events are in use that have callchains
670enabled, otherwise writing to this file will return -EBUSY.
671
672The default value is 127.
673
674==============================================================
675
676perf_event_max_contexts_per_stack:
677
678Controls maximum number of stack frame context entries for
679(attr.sample_type & PERF_SAMPLE_CALLCHAIN) configured events, for
680instance, when using 'perf record -g' or 'perf trace --call-graph fp'.
681
682This can only be done when no events are in use that have callchains
683enabled, otherwise writing to this file will return -EBUSY.
684
685The default value is 8.
686
687==============================================================
688
689pid_max:
690
691PID allocation wrap value. When the kernel's next PID value
692reaches this value, it wraps back to a minimum PID value.
693PIDs of value pid_max or larger are not allocated.
694
695==============================================================
696
697ns_last_pid:
698
699The last pid allocated in the current (the one task using this sysctl
700lives in) pid namespace. When selecting a pid for a next task on fork
701kernel tries to allocate a number starting from this one.
702
703==============================================================
704
705powersave-nap: (PPC only)
706
707If set, Linux-PPC will use the 'nap' mode of powersaving,
708otherwise the 'doze' mode will be used.
709
710==============================================================
711
712printk:
713
714The four values in printk denote: console_loglevel,
715default_message_loglevel, minimum_console_loglevel and
716default_console_loglevel respectively.
717
718These values influence printk() behavior when printing or
719logging error messages. See 'man 2 syslog' for more info on
720the different loglevels.
721
722- console_loglevel: messages with a higher priority than
723 this will be printed to the console
724- default_message_loglevel: messages without an explicit priority
725 will be printed with this priority
726- minimum_console_loglevel: minimum (highest) value to which
727 console_loglevel can be set
728- default_console_loglevel: default value for console_loglevel
729
730==============================================================
731
732printk_delay:
733
734Delay each printk message in printk_delay milliseconds
735
736Value from 0 - 10000 is allowed.
737
738==============================================================
739
740printk_ratelimit:
741
742Some warning messages are rate limited. printk_ratelimit specifies
743the minimum length of time between these messages (in jiffies), by
744default we allow one every 5 seconds.
745
746A value of 0 will disable rate limiting.
747
748==============================================================
749
750printk_ratelimit_burst:
751
752While long term we enforce one message per printk_ratelimit
753seconds, we do allow a burst of messages to pass through.
754printk_ratelimit_burst specifies the number of messages we can
755send before ratelimiting kicks in.
756
757==============================================================
758
759printk_devkmsg:
760
761Control the logging to /dev/kmsg from userspace:
762
763ratelimit: default, ratelimited
764on: unlimited logging to /dev/kmsg from userspace
765off: logging to /dev/kmsg disabled
766
767The kernel command line parameter printk.devkmsg= overrides this and is
768a one-time setting until next reboot: once set, it cannot be changed by
769this sysctl interface anymore.
770
771==============================================================
772
773randomize_va_space:
774
775This option can be used to select the type of process address
776space randomization that is used in the system, for architectures
777that support this feature.
778
7790 - Turn the process address space randomization off. This is the
780 default for architectures that do not support this feature anyways,
781 and kernels that are booted with the "norandmaps" parameter.
782
7831 - Make the addresses of mmap base, stack and VDSO page randomized.
784 This, among other things, implies that shared libraries will be
785 loaded to random addresses. Also for PIE-linked binaries, the
786 location of code start is randomized. This is the default if the
787 CONFIG_COMPAT_BRK option is enabled.
788
7892 - Additionally enable heap randomization. This is the default if
790 CONFIG_COMPAT_BRK is disabled.
791
792 There are a few legacy applications out there (such as some ancient
793 versions of libc.so.5 from 1996) that assume that brk area starts
794 just after the end of the code+bss. These applications break when
795 start of the brk area is randomized. There are however no known
796 non-legacy applications that would be broken this way, so for most
797 systems it is safe to choose full randomization.
798
799 Systems with ancient and/or broken binaries should be configured
800 with CONFIG_COMPAT_BRK enabled, which excludes the heap from process
801 address space randomization.
802
803==============================================================
804
805reboot-cmd: (Sparc only)
806
807??? This seems to be a way to give an argument to the Sparc
808ROM/Flash boot loader. Maybe to tell it what to do after
809rebooting. ???
810
811==============================================================
812
813rtsig-max & rtsig-nr:
814
815The file rtsig-max can be used to tune the maximum number
816of POSIX realtime (queued) signals that can be outstanding
817in the system.
818
819rtsig-nr shows the number of RT signals currently queued.
820
821==============================================================
822
823sched_schedstats:
824
825Enables/disables scheduler statistics. Enabling this feature
826incurs a small amount of overhead in the scheduler but is
827useful for debugging and performance tuning.
828
829==============================================================
830
831sg-big-buff:
832
833This file shows the size of the generic SCSI (sg) buffer.
834You can't tune it just yet, but you could change it on
835compile time by editing include/scsi/sg.h and changing
836the value of SG_BIG_BUFF.
837
838There shouldn't be any reason to change this value. If
839you can come up with one, you probably know what you
840are doing anyway :)
841
842==============================================================
843
844shmall:
845
846This parameter sets the total amount of shared memory pages that
847can be used system wide. Hence, SHMALL should always be at least
848ceil(shmmax/PAGE_SIZE).
849
850If you are not sure what the default PAGE_SIZE is on your Linux
851system, you can run the following command:
852
853# getconf PAGE_SIZE
854
855==============================================================
856
857shmmax:
858
859This value can be used to query and set the run time limit
860on the maximum shared memory segment size that can be created.
861Shared memory segments up to 1Gb are now supported in the
862kernel. This value defaults to SHMMAX.
863
864==============================================================
865
866shm_rmid_forced:
867
868Linux lets you set resource limits, including how much memory one
869process can consume, via setrlimit(2). Unfortunately, shared memory
870segments are allowed to exist without association with any process, and
871thus might not be counted against any resource limits. If enabled,
872shared memory segments are automatically destroyed when their attach
873count becomes zero after a detach or a process termination. It will
874also destroy segments that were created, but never attached to, on exit
875from the process. The only use left for IPC_RMID is to immediately
876destroy an unattached segment. Of course, this breaks the way things are
877defined, so some applications might stop working. Note that this
878feature will do you no good unless you also configure your resource
879limits (in particular, RLIMIT_AS and RLIMIT_NPROC). Most systems don't
880need this.
881
882Note that if you change this from 0 to 1, already created segments
883without users and with a dead originative process will be destroyed.
884
885==============================================================
886
887sysctl_writes_strict:
888
889Control how file position affects the behavior of updating sysctl values
890via the /proc/sys interface:
891
892 -1 - Legacy per-write sysctl value handling, with no printk warnings.
893 Each write syscall must fully contain the sysctl value to be
894 written, and multiple writes on the same sysctl file descriptor
895 will rewrite the sysctl value, regardless of file position.
896 0 - Same behavior as above, but warn about processes that perform writes
897 to a sysctl file descriptor when the file position is not 0.
898 1 - (default) Respect file position when writing sysctl strings. Multiple
899 writes will append to the sysctl value buffer. Anything past the max
900 length of the sysctl value buffer will be ignored. Writes to numeric
901 sysctl entries must always be at file position 0 and the value must
902 be fully contained in the buffer sent in the write syscall.
903
904==============================================================
905
906softlockup_all_cpu_backtrace:
907
908This value controls the soft lockup detector thread's behavior
909when a soft lockup condition is detected as to whether or not
910to gather further debug information. If enabled, each cpu will
911be issued an NMI and instructed to capture stack trace.
912
913This feature is only applicable for architectures which support
914NMI.
915
9160: do nothing. This is the default behavior.
917
9181: on detection capture more debug information.
919
920==============================================================
921
922soft_watchdog
923
924This parameter can be used to control the soft lockup detector.
925
926 0 - disable the soft lockup detector
927 1 - enable the soft lockup detector
928
929The soft lockup detector monitors CPUs for threads that are hogging the CPUs
930without rescheduling voluntarily, and thus prevent the 'watchdog/N' threads
931from running. The mechanism depends on the CPUs ability to respond to timer
932interrupts which are needed for the 'watchdog/N' threads to be woken up by
933the watchdog timer function, otherwise the NMI watchdog - if enabled - can
934detect a hard lockup condition.
935
936==============================================================
937
938tainted:
939
940Non-zero if the kernel has been tainted. Numeric values, which
941can be ORed together:
942
943 1 - A module with a non-GPL license has been loaded, this
944 includes modules with no license.
945 Set by modutils >= 2.4.9 and module-init-tools.
946 2 - A module was force loaded by insmod -f.
947 Set by modutils >= 2.4.9 and module-init-tools.
948 4 - Unsafe SMP processors: SMP with CPUs not designed for SMP.
949 8 - A module was forcibly unloaded from the system by rmmod -f.
950 16 - A hardware machine check error occurred on the system.
951 32 - A bad page was discovered on the system.
952 64 - The user has asked that the system be marked "tainted". This
953 could be because they are running software that directly modifies
954 the hardware, or for other reasons.
955 128 - The system has died.
956 256 - The ACPI DSDT has been overridden with one supplied by the user
957 instead of using the one provided by the hardware.
958 512 - A kernel warning has occurred.
9591024 - A module from drivers/staging was loaded.
9602048 - The system is working around a severe firmware bug.
9614096 - An out-of-tree module has been loaded.
9628192 - An unsigned module has been loaded in a kernel supporting module
963 signature.
96416384 - A soft lockup has previously occurred on the system.
96532768 - The kernel has been live patched.
966
967==============================================================
968
969threads-max
970
971This value controls the maximum number of threads that can be created
972using fork().
973
974During initialization the kernel sets this value such that even if the
975maximum number of threads is created, the thread structures occupy only
976a part (1/8th) of the available RAM pages.
977
978The minimum value that can be written to threads-max is 20.
979The maximum value that can be written to threads-max is given by the
980constant FUTEX_TID_MASK (0x3fffffff).
981If a value outside of this range is written to threads-max an error
982EINVAL occurs.
983
984The value written is checked against the available RAM pages. If the
985thread structures would occupy too much (more than 1/8th) of the
986available RAM pages threads-max is reduced accordingly.
987
988==============================================================
989
990unknown_nmi_panic:
991
992The value in this file affects behavior of handling NMI. When the
993value is non-zero, unknown NMI is trapped and then panic occurs. At
994that time, kernel debugging information is displayed on console.
995
996NMI switch that most IA32 servers have fires unknown NMI up, for
997example. If a system hangs up, try pressing the NMI switch.
998
999==============================================================
1000
1001watchdog:
1002
1003This parameter can be used to disable or enable the soft lockup detector
1004_and_ the NMI watchdog (i.e. the hard lockup detector) at the same time.
1005
1006 0 - disable both lockup detectors
1007 1 - enable both lockup detectors
1008
1009The soft lockup detector and the NMI watchdog can also be disabled or
1010enabled individually, using the soft_watchdog and nmi_watchdog parameters.
1011If the watchdog parameter is read, for example by executing
1012
1013 cat /proc/sys/kernel/watchdog
1014
1015the output of this command (0 or 1) shows the logical OR of soft_watchdog
1016and nmi_watchdog.
1017
1018==============================================================
1019
1020watchdog_cpumask:
1021
1022This value can be used to control on which cpus the watchdog may run.
1023The default cpumask is all possible cores, but if NO_HZ_FULL is
1024enabled in the kernel config, and cores are specified with the
1025nohz_full= boot argument, those cores are excluded by default.
1026Offline cores can be included in this mask, and if the core is later
1027brought online, the watchdog will be started based on the mask value.
1028
1029Typically this value would only be touched in the nohz_full case
1030to re-enable cores that by default were not running the watchdog,
1031if a kernel lockup was suspected on those cores.
1032
1033The argument value is the standard cpulist format for cpumasks,
1034so for example to enable the watchdog on cores 0, 2, 3, and 4 you
1035might say:
1036
1037 echo 0,2-4 > /proc/sys/kernel/watchdog_cpumask
1038
1039==============================================================
1040
1041watchdog_thresh:
1042
1043This value can be used to control the frequency of hrtimer and NMI
1044events and the soft and hard lockup thresholds. The default threshold
1045is 10 seconds.
1046
1047The softlockup threshold is (2 * watchdog_thresh). Setting this
1048tunable to zero will disable lockup detection altogether.
1049
1050==============================================================