Loading...
1#
2# Timer subsystem related configuration options
3#
4
5# Options selectable by arch Kconfig
6
7# Watchdog function for clocksources to detect instabilities
8config CLOCKSOURCE_WATCHDOG
9 bool
10
11# Architecture has extra clocksource data
12config ARCH_CLOCKSOURCE_DATA
13 bool
14
15# Timekeeping vsyscall support
16config GENERIC_TIME_VSYSCALL
17 bool
18
19# Timekeeping vsyscall support
20config GENERIC_TIME_VSYSCALL_OLD
21 bool
22
23# ktime_t scalar 64bit nsec representation
24config KTIME_SCALAR
25 bool
26
27# Old style timekeeping
28config ARCH_USES_GETTIMEOFFSET
29 bool
30
31# The generic clock events infrastructure
32config GENERIC_CLOCKEVENTS
33 bool
34
35# Migration helper. Builds, but does not invoke
36config GENERIC_CLOCKEVENTS_BUILD
37 bool
38 default y
39 depends on GENERIC_CLOCKEVENTS
40
41# Architecture can handle broadcast in a driver-agnostic way
42config ARCH_HAS_TICK_BROADCAST
43 bool
44
45# Clockevents broadcasting infrastructure
46config GENERIC_CLOCKEVENTS_BROADCAST
47 bool
48 depends on GENERIC_CLOCKEVENTS
49
50# Automatically adjust the min. reprogramming time for
51# clock event device
52config GENERIC_CLOCKEVENTS_MIN_ADJUST
53 bool
54
55# Generic update of CMOS clock
56config GENERIC_CMOS_UPDATE
57 bool
58
59if GENERIC_CLOCKEVENTS
60menu "Timers subsystem"
61
62# Core internal switch. Selected by NO_HZ_COMMON / HIGH_RES_TIMERS. This is
63# only related to the tick functionality. Oneshot clockevent devices
64# are supported independ of this.
65config TICK_ONESHOT
66 bool
67
68config NO_HZ_COMMON
69 bool
70 depends on !ARCH_USES_GETTIMEOFFSET && GENERIC_CLOCKEVENTS
71 select TICK_ONESHOT
72
73choice
74 prompt "Timer tick handling"
75 default NO_HZ_IDLE if NO_HZ
76
77config HZ_PERIODIC
78 bool "Periodic timer ticks (constant rate, no dynticks)"
79 help
80 This option keeps the tick running periodically at a constant
81 rate, even when the CPU doesn't need it.
82
83config NO_HZ_IDLE
84 bool "Idle dynticks system (tickless idle)"
85 depends on !ARCH_USES_GETTIMEOFFSET && GENERIC_CLOCKEVENTS
86 select NO_HZ_COMMON
87 help
88 This option enables a tickless idle system: timer interrupts
89 will only trigger on an as-needed basis when the system is idle.
90 This is usually interesting for energy saving.
91
92 Most of the time you want to say Y here.
93
94config NO_HZ_FULL
95 bool "Full dynticks system (tickless)"
96 # NO_HZ_COMMON dependency
97 depends on !ARCH_USES_GETTIMEOFFSET && GENERIC_CLOCKEVENTS
98 # We need at least one periodic CPU for timekeeping
99 depends on SMP
100 # RCU_USER_QS dependency
101 depends on HAVE_CONTEXT_TRACKING
102 # VIRT_CPU_ACCOUNTING_GEN dependency
103 depends on HAVE_VIRT_CPU_ACCOUNTING_GEN
104 select NO_HZ_COMMON
105 select RCU_USER_QS
106 select RCU_NOCB_CPU
107 select VIRT_CPU_ACCOUNTING_GEN
108 select IRQ_WORK
109 help
110 Adaptively try to shutdown the tick whenever possible, even when
111 the CPU is running tasks. Typically this requires running a single
112 task on the CPU. Chances for running tickless are maximized when
113 the task mostly runs in userspace and has few kernel activity.
114
115 You need to fill up the nohz_full boot parameter with the
116 desired range of dynticks CPUs.
117
118 This is implemented at the expense of some overhead in user <-> kernel
119 transitions: syscalls, exceptions and interrupts. Even when it's
120 dynamically off.
121
122 Say N.
123
124endchoice
125
126config NO_HZ_FULL_ALL
127 bool "Full dynticks system on all CPUs by default (except CPU 0)"
128 depends on NO_HZ_FULL
129 help
130 If the user doesn't pass the nohz_full boot option to
131 define the range of full dynticks CPUs, consider that all
132 CPUs in the system are full dynticks by default.
133 Note the boot CPU will still be kept outside the range to
134 handle the timekeeping duty.
135
136config NO_HZ_FULL_SYSIDLE
137 bool "Detect full-system idle state for full dynticks system"
138 depends on NO_HZ_FULL
139 default n
140 help
141 At least one CPU must keep the scheduling-clock tick running for
142 timekeeping purposes whenever there is a non-idle CPU, where
143 "non-idle" also includes dynticks CPUs as long as they are
144 running non-idle tasks. Because the underlying adaptive-tick
145 support cannot distinguish between all CPUs being idle and
146 all CPUs each running a single task in dynticks mode, the
147 underlying support simply ensures that there is always a CPU
148 handling the scheduling-clock tick, whether or not all CPUs
149 are idle. This Kconfig option enables scalable detection of
150 the all-CPUs-idle state, thus allowing the scheduling-clock
151 tick to be disabled when all CPUs are idle. Note that scalable
152 detection of the all-CPUs-idle state means that larger systems
153 will be slower to declare the all-CPUs-idle state.
154
155 Say Y if you would like to help debug all-CPUs-idle detection.
156
157 Say N if you are unsure.
158
159config NO_HZ_FULL_SYSIDLE_SMALL
160 int "Number of CPUs above which large-system approach is used"
161 depends on NO_HZ_FULL_SYSIDLE
162 range 1 NR_CPUS
163 default 8
164 help
165 The full-system idle detection mechanism takes a lazy approach
166 on large systems, as is required to attain decent scalability.
167 However, on smaller systems, scalability is not anywhere near as
168 large a concern as is energy efficiency. The sysidle subsystem
169 therefore uses a fast but non-scalable algorithm for small
170 systems and a lazier but scalable algorithm for large systems.
171 This Kconfig parameter defines the number of CPUs in the largest
172 system that will be considered to be "small".
173
174 The default value will be fine in most cases. Battery-powered
175 systems that (1) enable NO_HZ_FULL_SYSIDLE, (2) have larger
176 numbers of CPUs, and (3) are suffering from battery-lifetime
177 problems due to long sysidle latencies might wish to experiment
178 with larger values for this Kconfig parameter. On the other
179 hand, they might be even better served by disabling NO_HZ_FULL
180 entirely, given that NO_HZ_FULL is intended for HPC and
181 real-time workloads that at present do not tend to be run on
182 battery-powered systems.
183
184 Take the default if you are unsure.
185
186config NO_HZ
187 bool "Old Idle dynticks config"
188 depends on !ARCH_USES_GETTIMEOFFSET && GENERIC_CLOCKEVENTS
189 help
190 This is the old config entry that enables dynticks idle.
191 We keep it around for a little while to enforce backward
192 compatibility with older config files.
193
194config HIGH_RES_TIMERS
195 bool "High Resolution Timer Support"
196 depends on !ARCH_USES_GETTIMEOFFSET && GENERIC_CLOCKEVENTS
197 select TICK_ONESHOT
198 help
199 This option enables high resolution timer support. If your
200 hardware is not capable then this option only increases
201 the size of the kernel image.
202
203endmenu
204endif
1# SPDX-License-Identifier: GPL-2.0-only
2#
3# Timer subsystem related configuration options
4#
5
6# Options selectable by arch Kconfig
7
8# Watchdog function for clocksources to detect instabilities
9config CLOCKSOURCE_WATCHDOG
10 bool
11
12# Architecture has extra clocksource data
13config ARCH_CLOCKSOURCE_DATA
14 bool
15
16# Architecture has extra clocksource init called from registration
17config ARCH_CLOCKSOURCE_INIT
18 bool
19
20# Clocksources require validation of the clocksource against the last
21# cycle update - x86/TSC misfeature
22config CLOCKSOURCE_VALIDATE_LAST_CYCLE
23 bool
24
25# Timekeeping vsyscall support
26config GENERIC_TIME_VSYSCALL
27 bool
28
29# The generic clock events infrastructure
30config GENERIC_CLOCKEVENTS
31 def_bool !LEGACY_TIMER_TICK
32
33# Architecture can handle broadcast in a driver-agnostic way
34config ARCH_HAS_TICK_BROADCAST
35 bool
36
37# Clockevents broadcasting infrastructure
38config GENERIC_CLOCKEVENTS_BROADCAST
39 bool
40 depends on GENERIC_CLOCKEVENTS
41
42# Handle broadcast in default_idle_call()
43config GENERIC_CLOCKEVENTS_BROADCAST_IDLE
44 bool
45 depends on GENERIC_CLOCKEVENTS_BROADCAST
46
47# Automatically adjust the min. reprogramming time for
48# clock event device
49config GENERIC_CLOCKEVENTS_MIN_ADJUST
50 bool
51
52# Generic update of CMOS clock
53config GENERIC_CMOS_UPDATE
54 bool
55
56# Select to handle posix CPU timers from task_work
57# and not from the timer interrupt context
58config HAVE_POSIX_CPU_TIMERS_TASK_WORK
59 bool
60
61config POSIX_CPU_TIMERS_TASK_WORK
62 bool
63 default y if POSIX_TIMERS && HAVE_POSIX_CPU_TIMERS_TASK_WORK
64
65config LEGACY_TIMER_TICK
66 bool
67 help
68 The legacy timer tick helper is used by platforms that
69 lack support for the generic clockevent framework.
70 New platforms should use generic clockevents instead.
71
72config TIME_KUNIT_TEST
73 tristate "KUnit test for kernel/time functions" if !KUNIT_ALL_TESTS
74 depends on KUNIT
75 default KUNIT_ALL_TESTS
76 help
77 Enable this option to test RTC library functions.
78
79 If unsure, say N.
80
81config CONTEXT_TRACKING
82 bool
83
84config CONTEXT_TRACKING_IDLE
85 bool
86 select CONTEXT_TRACKING
87 help
88 Tracks idle state on behalf of RCU.
89
90if GENERIC_CLOCKEVENTS
91menu "Timers subsystem"
92
93# Core internal switch. Selected by NO_HZ_COMMON / HIGH_RES_TIMERS. This is
94# only related to the tick functionality. Oneshot clockevent devices
95# are supported independent of this.
96config TICK_ONESHOT
97 bool
98
99config NO_HZ_COMMON
100 bool
101 select TICK_ONESHOT
102
103choice
104 prompt "Timer tick handling"
105 default NO_HZ_IDLE if NO_HZ
106
107config HZ_PERIODIC
108 bool "Periodic timer ticks (constant rate, no dynticks)"
109 help
110 This option keeps the tick running periodically at a constant
111 rate, even when the CPU doesn't need it.
112
113config NO_HZ_IDLE
114 bool "Idle dynticks system (tickless idle)"
115 select NO_HZ_COMMON
116 help
117 This option enables a tickless idle system: timer interrupts
118 will only trigger on an as-needed basis when the system is idle.
119 This is usually interesting for energy saving.
120
121 Most of the time you want to say Y here.
122
123config NO_HZ_FULL
124 bool "Full dynticks system (tickless)"
125 # NO_HZ_COMMON dependency
126 # We need at least one periodic CPU for timekeeping
127 depends on SMP
128 depends on HAVE_CONTEXT_TRACKING_USER
129 # VIRT_CPU_ACCOUNTING_GEN dependency
130 depends on HAVE_VIRT_CPU_ACCOUNTING_GEN
131 select NO_HZ_COMMON
132 select RCU_NOCB_CPU
133 select VIRT_CPU_ACCOUNTING_GEN
134 select IRQ_WORK
135 select CPU_ISOLATION
136 help
137 Adaptively try to shutdown the tick whenever possible, even when
138 the CPU is running tasks. Typically this requires running a single
139 task on the CPU. Chances for running tickless are maximized when
140 the task mostly runs in userspace and has few kernel activity.
141
142 You need to fill up the nohz_full boot parameter with the
143 desired range of dynticks CPUs to use it. This is implemented at
144 the expense of some overhead in user <-> kernel transitions:
145 syscalls, exceptions and interrupts.
146
147 By default, without passing the nohz_full parameter, this behaves just
148 like NO_HZ_IDLE.
149
150 If you're a distro say Y.
151
152endchoice
153
154config CONTEXT_TRACKING_USER
155 bool
156 depends on HAVE_CONTEXT_TRACKING_USER
157 select CONTEXT_TRACKING
158 help
159 Track transitions between kernel and user on behalf of RCU and
160 tickless cputime accounting. The former case relies on context
161 tracking to enter/exit RCU extended quiescent states.
162
163config CONTEXT_TRACKING_USER_FORCE
164 bool "Force user context tracking"
165 depends on CONTEXT_TRACKING_USER
166 default y if !NO_HZ_FULL
167 help
168 The major pre-requirement for full dynticks to work is to
169 support the user context tracking subsystem. But there are also
170 other dependencies to provide in order to make the full
171 dynticks working.
172
173 This option stands for testing when an arch implements the
174 user context tracking backend but doesn't yet fulfill all the
175 requirements to make the full dynticks feature working.
176 Without the full dynticks, there is no way to test the support
177 for user context tracking and the subsystems that rely on it: RCU
178 userspace extended quiescent state and tickless cputime
179 accounting. This option copes with the absence of the full
180 dynticks subsystem by forcing the user context tracking on all
181 CPUs in the system.
182
183 Say Y only if you're working on the development of an
184 architecture backend for the user context tracking.
185
186 Say N otherwise, this option brings an overhead that you
187 don't want in production.
188
189config NO_HZ
190 bool "Old Idle dynticks config"
191 help
192 This is the old config entry that enables dynticks idle.
193 We keep it around for a little while to enforce backward
194 compatibility with older config files.
195
196config HIGH_RES_TIMERS
197 bool "High Resolution Timer Support"
198 select TICK_ONESHOT
199 help
200 This option enables high resolution timer support. If your
201 hardware is not capable then this option only increases
202 the size of the kernel image.
203
204config CLOCKSOURCE_WATCHDOG_MAX_SKEW_US
205 int "Clocksource watchdog maximum allowable skew (in μs)"
206 depends on CLOCKSOURCE_WATCHDOG
207 range 50 1000
208 default 125
209 help
210 Specify the maximum amount of allowable watchdog skew in
211 microseconds before reporting the clocksource to be unstable.
212 The default is based on a half-second clocksource watchdog
213 interval and NTP's maximum frequency drift of 500 parts
214 per million. If the clocksource is good enough for NTP,
215 it is good enough for the clocksource watchdog!
216
217endmenu
218endif