Loading...
1# SPDX-License-Identifier: GPL-2.0-only
2#
3# RCU-related configuration options
4#
5
6menu "RCU Subsystem"
7
8config TREE_RCU
9 bool
10 default y if SMP
11 # Dynticks-idle tracking
12 select CONTEXT_TRACKING_IDLE
13 help
14 This option selects the RCU implementation that is
15 designed for very large SMP system with hundreds or
16 thousands of CPUs. It also scales down nicely to
17 smaller systems.
18
19config PREEMPT_RCU
20 bool
21 default y if PREEMPTION
22 select TREE_RCU
23 help
24 This option selects the RCU implementation that is
25 designed for very large SMP systems with hundreds or
26 thousands of CPUs, but for which real-time response
27 is also required. It also scales down nicely to
28 smaller systems.
29
30 Select this option if you are unsure.
31
32config TINY_RCU
33 bool
34 default y if !PREEMPTION && !SMP
35 help
36 This option selects the RCU implementation that is
37 designed for UP systems from which real-time response
38 is not required. This option greatly reduces the
39 memory footprint of RCU.
40
41config RCU_EXPERT
42 bool "Make expert-level adjustments to RCU configuration"
43 default n
44 help
45 This option needs to be enabled if you wish to make
46 expert-level adjustments to RCU configuration. By default,
47 no such adjustments can be made, which has the often-beneficial
48 side-effect of preventing "make oldconfig" from asking you all
49 sorts of detailed questions about how you would like numerous
50 obscure RCU options to be set up.
51
52 Say Y if you need to make expert-level adjustments to RCU.
53
54 Say N if you are unsure.
55
56config TINY_SRCU
57 bool
58 default y if TINY_RCU
59 help
60 This option selects the single-CPU non-preemptible version of SRCU.
61
62config TREE_SRCU
63 bool
64 default y if !TINY_RCU
65 help
66 This option selects the full-fledged version of SRCU.
67
68config NEED_SRCU_NMI_SAFE
69 def_bool HAVE_NMI && !ARCH_HAS_NMI_SAFE_THIS_CPU_OPS && !TINY_SRCU
70
71config TASKS_RCU_GENERIC
72 def_bool TASKS_RCU || TASKS_RUDE_RCU || TASKS_TRACE_RCU
73 help
74 This option enables generic infrastructure code supporting
75 task-based RCU implementations. Not for manual selection.
76
77config FORCE_TASKS_RCU
78 bool "Force selection of TASKS_RCU"
79 depends on RCU_EXPERT
80 select TASKS_RCU
81 default n
82 help
83 This option force-enables a task-based RCU implementation
84 that uses only voluntary context switch (not preemption!),
85 idle, and user-mode execution as quiescent states. Not for
86 manual selection in most cases.
87
88config TASKS_RCU
89 bool
90 default n
91 select IRQ_WORK
92
93config FORCE_TASKS_RUDE_RCU
94 bool "Force selection of Tasks Rude RCU"
95 depends on RCU_EXPERT
96 select TASKS_RUDE_RCU
97 default n
98 help
99 This option force-enables a task-based RCU implementation
100 that uses only context switch (including preemption) and
101 user-mode execution as quiescent states. It forces IPIs and
102 context switches on all online CPUs, including idle ones,
103 so use with caution. Not for manual selection in most cases.
104
105config TASKS_RUDE_RCU
106 bool
107 default n
108 select IRQ_WORK
109
110config FORCE_TASKS_TRACE_RCU
111 bool "Force selection of Tasks Trace RCU"
112 depends on RCU_EXPERT
113 select TASKS_TRACE_RCU
114 default n
115 help
116 This option enables a task-based RCU implementation that uses
117 explicit rcu_read_lock_trace() read-side markers, and allows
118 these readers to appear in the idle loop as well as on the
119 CPU hotplug code paths. It can force IPIs on online CPUs,
120 including idle ones, so use with caution. Not for manual
121 selection in most cases.
122
123config TASKS_TRACE_RCU
124 bool
125 default n
126 select IRQ_WORK
127
128config RCU_STALL_COMMON
129 def_bool TREE_RCU
130 help
131 This option enables RCU CPU stall code that is common between
132 the TINY and TREE variants of RCU. The purpose is to allow
133 the tiny variants to disable RCU CPU stall warnings, while
134 making these warnings mandatory for the tree variants.
135
136config RCU_NEED_SEGCBLIST
137 def_bool ( TREE_RCU || TREE_SRCU || TASKS_RCU_GENERIC )
138
139config RCU_FANOUT
140 int "Tree-based hierarchical RCU fanout value"
141 range 2 64 if 64BIT
142 range 2 32 if !64BIT
143 depends on TREE_RCU && RCU_EXPERT
144 default 64 if 64BIT
145 default 32 if !64BIT
146 help
147 This option controls the fanout of hierarchical implementations
148 of RCU, allowing RCU to work efficiently on machines with
149 large numbers of CPUs. This value must be at least the fourth
150 root of NR_CPUS, which allows NR_CPUS to be insanely large.
151 The default value of RCU_FANOUT should be used for production
152 systems, but if you are stress-testing the RCU implementation
153 itself, small RCU_FANOUT values allow you to test large-system
154 code paths on small(er) systems.
155
156 Select a specific number if testing RCU itself.
157 Take the default if unsure.
158
159config RCU_FANOUT_LEAF
160 int "Tree-based hierarchical RCU leaf-level fanout value"
161 range 2 64 if 64BIT && !RCU_STRICT_GRACE_PERIOD
162 range 2 32 if !64BIT && !RCU_STRICT_GRACE_PERIOD
163 range 2 3 if RCU_STRICT_GRACE_PERIOD
164 depends on TREE_RCU && RCU_EXPERT
165 default 16 if !RCU_STRICT_GRACE_PERIOD
166 default 2 if RCU_STRICT_GRACE_PERIOD
167 help
168 This option controls the leaf-level fanout of hierarchical
169 implementations of RCU, and allows trading off cache misses
170 against lock contention. Systems that synchronize their
171 scheduling-clock interrupts for energy-efficiency reasons will
172 want the default because the smaller leaf-level fanout keeps
173 lock contention levels acceptably low. Very large systems
174 (hundreds or thousands of CPUs) will instead want to set this
175 value to the maximum value possible in order to reduce the
176 number of cache misses incurred during RCU's grace-period
177 initialization. These systems tend to run CPU-bound, and thus
178 are not helped by synchronized interrupts, and thus tend to
179 skew them, which reduces lock contention enough that large
180 leaf-level fanouts work well. That said, setting leaf-level
181 fanout to a large number will likely cause problematic
182 lock contention on the leaf-level rcu_node structures unless
183 you boot with the skew_tick kernel parameter.
184
185 Select a specific number if testing RCU itself.
186
187 Select the maximum permissible value for large systems, but
188 please understand that you may also need to set the skew_tick
189 kernel boot parameter to avoid contention on the rcu_node
190 structure's locks.
191
192 Take the default if unsure.
193
194config RCU_BOOST
195 bool "Enable RCU priority boosting"
196 depends on (RT_MUTEXES && PREEMPT_RCU && RCU_EXPERT) || PREEMPT_RT
197 default y if PREEMPT_RT
198 help
199 This option boosts the priority of preempted RCU readers that
200 block the current preemptible RCU grace period for too long.
201 This option also prevents heavy loads from blocking RCU
202 callback invocation.
203
204 Say Y here if you are working with real-time apps or heavy loads
205 Say N here if you are unsure.
206
207config RCU_BOOST_DELAY
208 int "Milliseconds to delay boosting after RCU grace-period start"
209 range 0 3000
210 depends on RCU_BOOST
211 default 500
212 help
213 This option specifies the time to wait after the beginning of
214 a given grace period before priority-boosting preempted RCU
215 readers blocking that grace period. Note that any RCU reader
216 blocking an expedited RCU grace period is boosted immediately.
217
218 Accept the default if unsure.
219
220config RCU_EXP_KTHREAD
221 bool "Perform RCU expedited work in a real-time kthread"
222 depends on RCU_BOOST && RCU_EXPERT
223 default !PREEMPT_RT && NR_CPUS <= 32
224 help
225 Use this option to further reduce the latencies of expedited
226 grace periods at the expense of being more disruptive.
227
228 This option is disabled by default on PREEMPT_RT=y kernels which
229 disable expedited grace periods after boot by unconditionally
230 setting rcupdate.rcu_normal_after_boot=1.
231
232 Accept the default if unsure.
233
234config RCU_NOCB_CPU
235 bool "Offload RCU callback processing from boot-selected CPUs"
236 depends on TREE_RCU
237 depends on RCU_EXPERT || NO_HZ_FULL
238 default n
239 help
240 Use this option to reduce OS jitter for aggressive HPC or
241 real-time workloads. It can also be used to offload RCU
242 callback invocation to energy-efficient CPUs in battery-powered
243 asymmetric multiprocessors. The price of this reduced jitter
244 is that the overhead of call_rcu() increases and that some
245 workloads will incur significant increases in context-switch
246 rates.
247
248 This option offloads callback invocation from the set of CPUs
249 specified at boot time by the rcu_nocbs parameter. For each
250 such CPU, a kthread ("rcuox/N") will be created to invoke
251 callbacks, where the "N" is the CPU being offloaded, and where
252 the "x" is "p" for RCU-preempt (PREEMPTION kernels) and "s" for
253 RCU-sched (!PREEMPTION kernels). Nothing prevents this kthread
254 from running on the specified CPUs, but (1) the kthreads may be
255 preempted between each callback, and (2) affinity or cgroups can
256 be used to force the kthreads to run on whatever set of CPUs is
257 desired.
258
259 Say Y here if you need reduced OS jitter, despite added overhead.
260 Say N here if you are unsure.
261
262config RCU_NOCB_CPU_DEFAULT_ALL
263 bool "Offload RCU callback processing from all CPUs by default"
264 depends on RCU_NOCB_CPU
265 default n
266 help
267 Use this option to offload callback processing from all CPUs
268 by default, in the absence of the rcu_nocbs or nohz_full boot
269 parameter. This also avoids the need to use any boot parameters
270 to achieve the effect of offloading all CPUs on boot.
271
272 Say Y here if you want offload all CPUs by default on boot.
273 Say N here if you are unsure.
274
275config RCU_NOCB_CPU_CB_BOOST
276 bool "Offload RCU callback from real-time kthread"
277 depends on RCU_NOCB_CPU && RCU_BOOST
278 default y if PREEMPT_RT
279 help
280 Use this option to invoke offloaded callbacks as SCHED_FIFO
281 to avoid starvation by heavy SCHED_OTHER background load.
282 Of course, running as SCHED_FIFO during callback floods will
283 cause the rcuo[ps] kthreads to monopolize the CPU for hundreds
284 of milliseconds or more. Therefore, when enabling this option,
285 it is your responsibility to ensure that latency-sensitive
286 tasks either run with higher priority or run on some other CPU.
287
288 Say Y here if you want to set RT priority for offloading kthreads.
289 Say N here if you are building a !PREEMPT_RT kernel and are unsure.
290
291config TASKS_TRACE_RCU_READ_MB
292 bool "Tasks Trace RCU readers use memory barriers in user and idle"
293 depends on RCU_EXPERT && TASKS_TRACE_RCU
294 default PREEMPT_RT || NR_CPUS < 8
295 help
296 Use this option to further reduce the number of IPIs sent
297 to CPUs executing in userspace or idle during tasks trace
298 RCU grace periods. Given that a reasonable setting of
299 the rcupdate.rcu_task_ipi_delay kernel boot parameter
300 eliminates such IPIs for many workloads, proper setting
301 of this Kconfig option is important mostly for aggressive
302 real-time installations and for battery-powered devices,
303 hence the default chosen above.
304
305 Say Y here if you hate IPIs.
306 Say N here if you hate read-side memory barriers.
307 Take the default if you are unsure.
308
309config RCU_LAZY
310 bool "RCU callback lazy invocation functionality"
311 depends on RCU_NOCB_CPU
312 default n
313 help
314 To save power, batch RCU callbacks and flush after delay, memory
315 pressure, or callback list growing too big.
316
317 Requires rcu_nocbs=all to be set.
318
319 Use rcutree.enable_rcu_lazy=0 to turn it off at boot time.
320
321config RCU_LAZY_DEFAULT_OFF
322 bool "Turn RCU lazy invocation off by default"
323 depends on RCU_LAZY
324 default n
325 help
326 Allows building the kernel with CONFIG_RCU_LAZY=y yet keep it default
327 off. Boot time param rcutree.enable_rcu_lazy=1 can be used to switch
328 it back on.
329
330config RCU_DOUBLE_CHECK_CB_TIME
331 bool "RCU callback-batch backup time check"
332 depends on RCU_EXPERT
333 default n
334 help
335 Use this option to provide more precise enforcement of the
336 rcutree.rcu_resched_ns module parameter in situations where
337 a single RCU callback might run for hundreds of microseconds,
338 thus defeating the 32-callback batching used to amortize the
339 cost of the fine-grained but expensive local_clock() function.
340
341 This option rounds rcutree.rcu_resched_ns up to the next
342 jiffy, and overrides the 32-callback batching if this limit
343 is exceeded.
344
345 Say Y here if you need tighter callback-limit enforcement.
346 Say N here if you are unsure.
347
348endmenu # "RCU Subsystem"
1# SPDX-License-Identifier: GPL-2.0-only
2#
3# RCU-related configuration options
4#
5
6menu "RCU Subsystem"
7
8config TREE_RCU
9 bool
10 default y if SMP
11 help
12 This option selects the RCU implementation that is
13 designed for very large SMP system with hundreds or
14 thousands of CPUs. It also scales down nicely to
15 smaller systems.
16
17config PREEMPT_RCU
18 bool
19 default y if PREEMPTION
20 select TREE_RCU
21 help
22 This option selects the RCU implementation that is
23 designed for very large SMP systems with hundreds or
24 thousands of CPUs, but for which real-time response
25 is also required. It also scales down nicely to
26 smaller systems.
27
28 Select this option if you are unsure.
29
30config TINY_RCU
31 bool
32 default y if !PREEMPTION && !SMP
33 help
34 This option selects the RCU implementation that is
35 designed for UP systems from which real-time response
36 is not required. This option greatly reduces the
37 memory footprint of RCU.
38
39config RCU_EXPERT
40 bool "Make expert-level adjustments to RCU configuration"
41 default n
42 help
43 This option needs to be enabled if you wish to make
44 expert-level adjustments to RCU configuration. By default,
45 no such adjustments can be made, which has the often-beneficial
46 side-effect of preventing "make oldconfig" from asking you all
47 sorts of detailed questions about how you would like numerous
48 obscure RCU options to be set up.
49
50 Say Y if you need to make expert-level adjustments to RCU.
51
52 Say N if you are unsure.
53
54config SRCU
55 bool
56 help
57 This option selects the sleepable version of RCU. This version
58 permits arbitrary sleeping or blocking within RCU read-side critical
59 sections.
60
61config TINY_SRCU
62 bool
63 default y if SRCU && TINY_RCU
64 help
65 This option selects the single-CPU non-preemptible version of SRCU.
66
67config TREE_SRCU
68 bool
69 default y if SRCU && !TINY_RCU
70 help
71 This option selects the full-fledged version of SRCU.
72
73config TASKS_RCU_GENERIC
74 def_bool TASKS_RCU || TASKS_RUDE_RCU || TASKS_TRACE_RCU
75 select SRCU
76 help
77 This option enables generic infrastructure code supporting
78 task-based RCU implementations. Not for manual selection.
79
80config TASKS_RCU
81 def_bool PREEMPTION
82 help
83 This option enables a task-based RCU implementation that uses
84 only voluntary context switch (not preemption!), idle, and
85 user-mode execution as quiescent states. Not for manual selection.
86
87config TASKS_RUDE_RCU
88 def_bool 0
89 help
90 This option enables a task-based RCU implementation that uses
91 only context switch (including preemption) and user-mode
92 execution as quiescent states. It forces IPIs and context
93 switches on all online CPUs, including idle ones, so use
94 with caution.
95
96config TASKS_TRACE_RCU
97 def_bool 0
98 select IRQ_WORK
99 help
100 This option enables a task-based RCU implementation that uses
101 explicit rcu_read_lock_trace() read-side markers, and allows
102 these readers to appear in the idle loop as well as on the CPU
103 hotplug code paths. It can force IPIs on online CPUs, including
104 idle ones, so use with caution.
105
106config RCU_STALL_COMMON
107 def_bool TREE_RCU
108 help
109 This option enables RCU CPU stall code that is common between
110 the TINY and TREE variants of RCU. The purpose is to allow
111 the tiny variants to disable RCU CPU stall warnings, while
112 making these warnings mandatory for the tree variants.
113
114config RCU_NEED_SEGCBLIST
115 def_bool ( TREE_RCU || TREE_SRCU )
116
117config RCU_FANOUT
118 int "Tree-based hierarchical RCU fanout value"
119 range 2 64 if 64BIT
120 range 2 32 if !64BIT
121 depends on TREE_RCU && RCU_EXPERT
122 default 64 if 64BIT
123 default 32 if !64BIT
124 help
125 This option controls the fanout of hierarchical implementations
126 of RCU, allowing RCU to work efficiently on machines with
127 large numbers of CPUs. This value must be at least the fourth
128 root of NR_CPUS, which allows NR_CPUS to be insanely large.
129 The default value of RCU_FANOUT should be used for production
130 systems, but if you are stress-testing the RCU implementation
131 itself, small RCU_FANOUT values allow you to test large-system
132 code paths on small(er) systems.
133
134 Select a specific number if testing RCU itself.
135 Take the default if unsure.
136
137config RCU_FANOUT_LEAF
138 int "Tree-based hierarchical RCU leaf-level fanout value"
139 range 2 64 if 64BIT && !RCU_STRICT_GRACE_PERIOD
140 range 2 32 if !64BIT && !RCU_STRICT_GRACE_PERIOD
141 range 2 3 if RCU_STRICT_GRACE_PERIOD
142 depends on TREE_RCU && RCU_EXPERT
143 default 16 if !RCU_STRICT_GRACE_PERIOD
144 default 2 if RCU_STRICT_GRACE_PERIOD
145 help
146 This option controls the leaf-level fanout of hierarchical
147 implementations of RCU, and allows trading off cache misses
148 against lock contention. Systems that synchronize their
149 scheduling-clock interrupts for energy-efficiency reasons will
150 want the default because the smaller leaf-level fanout keeps
151 lock contention levels acceptably low. Very large systems
152 (hundreds or thousands of CPUs) will instead want to set this
153 value to the maximum value possible in order to reduce the
154 number of cache misses incurred during RCU's grace-period
155 initialization. These systems tend to run CPU-bound, and thus
156 are not helped by synchronized interrupts, and thus tend to
157 skew them, which reduces lock contention enough that large
158 leaf-level fanouts work well. That said, setting leaf-level
159 fanout to a large number will likely cause problematic
160 lock contention on the leaf-level rcu_node structures unless
161 you boot with the skew_tick kernel parameter.
162
163 Select a specific number if testing RCU itself.
164
165 Select the maximum permissible value for large systems, but
166 please understand that you may also need to set the skew_tick
167 kernel boot parameter to avoid contention on the rcu_node
168 structure's locks.
169
170 Take the default if unsure.
171
172config RCU_FAST_NO_HZ
173 bool "Accelerate last non-dyntick-idle CPU's grace periods"
174 depends on NO_HZ_COMMON && SMP && RCU_EXPERT
175 default n
176 help
177 This option permits CPUs to enter dynticks-idle state even if
178 they have RCU callbacks queued, and prevents RCU from waking
179 these CPUs up more than roughly once every four jiffies (by
180 default, you can adjust this using the rcutree.rcu_idle_gp_delay
181 parameter), thus improving energy efficiency. On the other
182 hand, this option increases the duration of RCU grace periods,
183 for example, slowing down synchronize_rcu().
184
185 Say Y if energy efficiency is critically important, and you
186 don't care about increased grace-period durations.
187
188 Say N if you are unsure.
189
190config RCU_BOOST
191 bool "Enable RCU priority boosting"
192 depends on (RT_MUTEXES && PREEMPT_RCU && RCU_EXPERT) || PREEMPT_RT
193 default y if PREEMPT_RT
194 help
195 This option boosts the priority of preempted RCU readers that
196 block the current preemptible RCU grace period for too long.
197 This option also prevents heavy loads from blocking RCU
198 callback invocation.
199
200 Say Y here if you are working with real-time apps or heavy loads
201 Say N here if you are unsure.
202
203config RCU_BOOST_DELAY
204 int "Milliseconds to delay boosting after RCU grace-period start"
205 range 0 3000
206 depends on RCU_BOOST
207 default 500
208 help
209 This option specifies the time to wait after the beginning of
210 a given grace period before priority-boosting preempted RCU
211 readers blocking that grace period. Note that any RCU reader
212 blocking an expedited RCU grace period is boosted immediately.
213
214 Accept the default if unsure.
215
216config RCU_NOCB_CPU
217 bool "Offload RCU callback processing from boot-selected CPUs"
218 depends on TREE_RCU
219 depends on RCU_EXPERT || NO_HZ_FULL
220 default n
221 help
222 Use this option to reduce OS jitter for aggressive HPC or
223 real-time workloads. It can also be used to offload RCU
224 callback invocation to energy-efficient CPUs in battery-powered
225 asymmetric multiprocessors. The price of this reduced jitter
226 is that the overhead of call_rcu() increases and that some
227 workloads will incur significant increases in context-switch
228 rates.
229
230 This option offloads callback invocation from the set of CPUs
231 specified at boot time by the rcu_nocbs parameter. For each
232 such CPU, a kthread ("rcuox/N") will be created to invoke
233 callbacks, where the "N" is the CPU being offloaded, and where
234 the "x" is "p" for RCU-preempt (PREEMPTION kernels) and "s" for
235 RCU-sched (!PREEMPTION kernels). Nothing prevents this kthread
236 from running on the specified CPUs, but (1) the kthreads may be
237 preempted between each callback, and (2) affinity or cgroups can
238 be used to force the kthreads to run on whatever set of CPUs is
239 desired.
240
241 Say Y here if you need reduced OS jitter, despite added overhead.
242 Say N here if you are unsure.
243
244config TASKS_TRACE_RCU_READ_MB
245 bool "Tasks Trace RCU readers use memory barriers in user and idle"
246 depends on RCU_EXPERT
247 default PREEMPT_RT || NR_CPUS < 8
248 help
249 Use this option to further reduce the number of IPIs sent
250 to CPUs executing in userspace or idle during tasks trace
251 RCU grace periods. Given that a reasonable setting of
252 the rcupdate.rcu_task_ipi_delay kernel boot parameter
253 eliminates such IPIs for many workloads, proper setting
254 of this Kconfig option is important mostly for aggressive
255 real-time installations and for battery-powered devices,
256 hence the default chosen above.
257
258 Say Y here if you hate IPIs.
259 Say N here if you hate read-side memory barriers.
260 Take the default if you are unsure.
261
262endmenu # "RCU Subsystem"