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