1 # SPDX-License-Identifier: GPL-2.0-only
3 # RCU-related configuration options
11 # Dynticks-idle tracking
12 select CONTEXT_TRACKING_IDLE
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
21 default y if PREEMPTION
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
30 Select this option if you are unsure.
34 default y if !PREEMPT_RCU && !SMP
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.
42 bool "Make expert-level adjustments to RCU configuration"
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.
52 Say Y if you need to make expert-level adjustments to RCU.
54 Say N if you are unsure.
60 This option selects the single-CPU non-preemptible version of SRCU.
64 default y if !TINY_RCU
66 This option selects the full-fledged version of SRCU.
68 config NEED_SRCU_NMI_SAFE
69 def_bool HAVE_NMI && !ARCH_HAS_NMI_SAFE_THIS_CPU_OPS && !TINY_SRCU
71 config TASKS_RCU_GENERIC
72 def_bool TASKS_RCU || TASKS_RUDE_RCU || TASKS_TRACE_RCU
74 This option enables generic infrastructure code supporting
75 task-based RCU implementations. Not for manual selection.
77 config FORCE_TASKS_RCU
78 bool "Force selection of TASKS_RCU"
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.
94 default NEED_TASKS_RCU && (PREEMPTION || PREEMPT_AUTO)
97 config FORCE_TASKS_RUDE_RCU
98 bool "Force selection of Tasks Rude RCU"
100 select TASKS_RUDE_RCU
103 This option force-enables a task-based RCU implementation
104 that uses only context switch (including preemption) and
105 user-mode execution as quiescent states. It forces IPIs and
106 context switches on all online CPUs, including idle ones,
107 so use with caution. Not for manual selection in most cases.
109 config TASKS_RUDE_RCU
114 config FORCE_TASKS_TRACE_RCU
115 bool "Force selection of Tasks Trace RCU"
116 depends on RCU_EXPERT
117 select TASKS_TRACE_RCU
120 This option enables a task-based RCU implementation that uses
121 explicit rcu_read_lock_trace() read-side markers, and allows
122 these readers to appear in the idle loop as well as on the
123 CPU hotplug code paths. It can force IPIs on online CPUs,
124 including idle ones, so use with caution. Not for manual
125 selection in most cases.
127 config TASKS_TRACE_RCU
132 config RCU_STALL_COMMON
135 This option enables RCU CPU stall code that is common between
136 the TINY and TREE variants of RCU. The purpose is to allow
137 the tiny variants to disable RCU CPU stall warnings, while
138 making these warnings mandatory for the tree variants.
140 config RCU_NEED_SEGCBLIST
141 def_bool ( TREE_RCU || TREE_SRCU || TASKS_RCU_GENERIC )
144 int "Tree-based hierarchical RCU fanout value"
147 depends on TREE_RCU && RCU_EXPERT
151 This option controls the fanout of hierarchical implementations
152 of RCU, allowing RCU to work efficiently on machines with
153 large numbers of CPUs. This value must be at least the fourth
154 root of NR_CPUS, which allows NR_CPUS to be insanely large.
155 The default value of RCU_FANOUT should be used for production
156 systems, but if you are stress-testing the RCU implementation
157 itself, small RCU_FANOUT values allow you to test large-system
158 code paths on small(er) systems.
160 Select a specific number if testing RCU itself.
161 Take the default if unsure.
163 config RCU_FANOUT_LEAF
164 int "Tree-based hierarchical RCU leaf-level fanout value"
165 range 2 64 if 64BIT && !RCU_STRICT_GRACE_PERIOD
166 range 2 32 if !64BIT && !RCU_STRICT_GRACE_PERIOD
167 range 2 3 if RCU_STRICT_GRACE_PERIOD
168 depends on TREE_RCU && RCU_EXPERT
169 default 16 if !RCU_STRICT_GRACE_PERIOD
170 default 2 if RCU_STRICT_GRACE_PERIOD
172 This option controls the leaf-level fanout of hierarchical
173 implementations of RCU, and allows trading off cache misses
174 against lock contention. Systems that synchronize their
175 scheduling-clock interrupts for energy-efficiency reasons will
176 want the default because the smaller leaf-level fanout keeps
177 lock contention levels acceptably low. Very large systems
178 (hundreds or thousands of CPUs) will instead want to set this
179 value to the maximum value possible in order to reduce the
180 number of cache misses incurred during RCU's grace-period
181 initialization. These systems tend to run CPU-bound, and thus
182 are not helped by synchronized interrupts, and thus tend to
183 skew them, which reduces lock contention enough that large
184 leaf-level fanouts work well. That said, setting leaf-level
185 fanout to a large number will likely cause problematic
186 lock contention on the leaf-level rcu_node structures unless
187 you boot with the skew_tick kernel parameter.
189 Select a specific number if testing RCU itself.
191 Select the maximum permissible value for large systems, but
192 please understand that you may also need to set the skew_tick
193 kernel boot parameter to avoid contention on the rcu_node
196 Take the default if unsure.
199 bool "Enable RCU priority boosting"
200 depends on (RT_MUTEXES && PREEMPT_RCU && RCU_EXPERT) || PREEMPT_RT
201 default y if PREEMPT_RT
203 This option boosts the priority of preempted RCU readers that
204 block the current preemptible RCU grace period for too long.
205 This option also prevents heavy loads from blocking RCU
208 Say Y here if you are working with real-time apps or heavy loads
209 Say N here if you are unsure.
211 config RCU_BOOST_DELAY
212 int "Milliseconds to delay boosting after RCU grace-period start"
217 This option specifies the time to wait after the beginning of
218 a given grace period before priority-boosting preempted RCU
219 readers blocking that grace period. Note that any RCU reader
220 blocking an expedited RCU grace period is boosted immediately.
222 Accept the default if unsure.
224 config RCU_EXP_KTHREAD
225 bool "Perform RCU expedited work in a real-time kthread"
226 depends on RCU_BOOST && RCU_EXPERT
227 default !PREEMPT_RT && NR_CPUS <= 32
229 Use this option to further reduce the latencies of expedited
230 grace periods at the expense of being more disruptive.
232 This option is disabled by default on PREEMPT_RT=y kernels which
233 disable expedited grace periods after boot by unconditionally
234 setting rcupdate.rcu_normal_after_boot=1.
236 Accept the default if unsure.
239 bool "Offload RCU callback processing from boot-selected CPUs"
241 depends on RCU_EXPERT || NO_HZ_FULL
244 Use this option to reduce OS jitter for aggressive HPC or
245 real-time workloads. It can also be used to offload RCU
246 callback invocation to energy-efficient CPUs in battery-powered
247 asymmetric multiprocessors. The price of this reduced jitter
248 is that the overhead of call_rcu() increases and that some
249 workloads will incur significant increases in context-switch
252 This option offloads callback invocation from the set of
253 CPUs specified at boot time by the rcu_nocbs parameter.
254 For each such CPU, a kthread ("rcuox/N") will be created to
255 invoke callbacks, where the "N" is the CPU being offloaded,
256 and where the "x" is "p" for RCU-preempt (PREEMPTION kernels)
257 and "s" for RCU-sched (!PREEMPTION kernels). This option
258 also creates another kthread for each sqrt(nr_cpu_ids) CPUs
259 ("rcuog/N", where N is the first CPU in that group to come
260 online), which handles grace periods for its group. Nothing
261 prevents these kthreads from running on the specified CPUs,
262 but (1) the kthreads may be preempted between each callback,
263 and (2) affinity or cgroups can be used to force the kthreads
264 to run on whatever set of CPUs is desired.
266 The sqrt(nr_cpu_ids) grouping may be overridden using the
267 rcutree.rcu_nocb_gp_stride kernel boot parameter. This can
268 be especially helpful for smaller numbers of CPUs, where
269 sqrt(nr_cpu_ids) can be a bit of a blunt instrument.
271 Say Y here if you need reduced OS jitter, despite added overhead.
272 Say N here if you are unsure.
274 config RCU_NOCB_CPU_DEFAULT_ALL
275 bool "Offload RCU callback processing from all CPUs by default"
276 depends on RCU_NOCB_CPU
279 Use this option to offload callback processing from all CPUs
280 by default, in the absence of the rcu_nocbs or nohz_full boot
281 parameter. This also avoids the need to use any boot parameters
282 to achieve the effect of offloading all CPUs on boot.
284 Say Y here if you want offload all CPUs by default on boot.
285 Say N here if you are unsure.
287 config RCU_NOCB_CPU_CB_BOOST
288 bool "Offload RCU callback from real-time kthread"
289 depends on RCU_NOCB_CPU && RCU_BOOST
290 default y if PREEMPT_RT
292 Use this option to invoke offloaded callbacks as SCHED_FIFO
293 to avoid starvation by heavy SCHED_OTHER background load.
294 Of course, running as SCHED_FIFO during callback floods will
295 cause the rcuo[ps] kthreads to monopolize the CPU for hundreds
296 of milliseconds or more. Therefore, when enabling this option,
297 it is your responsibility to ensure that latency-sensitive
298 tasks either run with higher priority or run on some other CPU.
300 Say Y here if you want to set RT priority for offloading kthreads.
301 Say N here if you are building a !PREEMPT_RT kernel and are unsure.
303 config TASKS_TRACE_RCU_READ_MB
304 bool "Tasks Trace RCU readers use memory barriers in user and idle"
305 depends on RCU_EXPERT && TASKS_TRACE_RCU
306 default PREEMPT_RT || NR_CPUS < 8
308 Use this option to further reduce the number of IPIs sent
309 to CPUs executing in userspace or idle during tasks trace
310 RCU grace periods. Given that a reasonable setting of
311 the rcupdate.rcu_task_ipi_delay kernel boot parameter
312 eliminates such IPIs for many workloads, proper setting
313 of this Kconfig option is important mostly for aggressive
314 real-time installations and for battery-powered devices,
315 hence the default chosen above.
317 Say Y here if you hate IPIs.
318 Say N here if you hate read-side memory barriers.
319 Take the default if you are unsure.
322 bool "RCU callback lazy invocation functionality"
323 depends on RCU_NOCB_CPU
326 To save power, batch RCU callbacks and flush after delay, memory
327 pressure, or callback list growing too big.
329 Requires rcu_nocbs=all to be set.
331 Use rcutree.enable_rcu_lazy=0 to turn it off at boot time.
333 config RCU_LAZY_DEFAULT_OFF
334 bool "Turn RCU lazy invocation off by default"
338 Allows building the kernel with CONFIG_RCU_LAZY=y yet keep it default
339 off. Boot time param rcutree.enable_rcu_lazy=1 can be used to switch
342 config RCU_DOUBLE_CHECK_CB_TIME
343 bool "RCU callback-batch backup time check"
344 depends on RCU_EXPERT
347 Use this option to provide more precise enforcement of the
348 rcutree.rcu_resched_ns module parameter in situations where
349 a single RCU callback might run for hundreds of microseconds,
350 thus defeating the 32-callback batching used to amortize the
351 cost of the fine-grained but expensive local_clock() function.
353 This option rounds rcutree.rcu_resched_ns up to the next
354 jiffy, and overrides the 32-callback batching if this limit
357 Say Y here if you need tighter callback-limit enforcement.
358 Say N here if you are unsure.
360 endmenu # "RCU Subsystem"