xref: /linux/kernel/rcu/Kconfig (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 !PREEMPT_RCU && !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 NEED_TASKS_RCU
89	bool
90	default n
91
92config TASKS_RCU
93	bool
94	default NEED_TASKS_RCU && (PREEMPTION || PREEMPT_AUTO)
95	select IRQ_WORK
96
97config FORCE_TASKS_RUDE_RCU
98	bool "Force selection of Tasks Rude RCU"
99	depends on RCU_EXPERT
100	select TASKS_RUDE_RCU
101	default n
102	help
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.
108
109config TASKS_RUDE_RCU
110	bool
111	default n
112	select IRQ_WORK
113
114config FORCE_TASKS_TRACE_RCU
115	bool "Force selection of Tasks Trace RCU"
116	depends on RCU_EXPERT
117	select TASKS_TRACE_RCU
118	default n
119	help
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.
126
127config TASKS_TRACE_RCU
128	bool
129	default n
130	select IRQ_WORK
131
132config RCU_STALL_COMMON
133	def_bool TREE_RCU
134	help
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.
139
140config RCU_NEED_SEGCBLIST
141	def_bool ( TREE_RCU || TREE_SRCU || TASKS_RCU_GENERIC )
142
143config RCU_FANOUT
144	int "Tree-based hierarchical RCU fanout value"
145	range 2 64 if 64BIT
146	range 2 32 if !64BIT
147	depends on TREE_RCU && RCU_EXPERT
148	default 64 if 64BIT
149	default 32 if !64BIT
150	help
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.
159
160	  Select a specific number if testing RCU itself.
161	  Take the default if unsure.
162
163config 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
171	help
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.
188
189	  Select a specific number if testing RCU itself.
190
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
194	  structure's locks.
195
196	  Take the default if unsure.
197
198config RCU_BOOST
199	bool "Enable RCU priority boosting"
200	depends on (RT_MUTEXES && PREEMPT_RCU && RCU_EXPERT) || PREEMPT_RT
201	default y if PREEMPT_RT
202	help
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
206	  callback invocation.
207
208	  Say Y here if you are working with real-time apps or heavy loads
209	  Say N here if you are unsure.
210
211config RCU_BOOST_DELAY
212	int "Milliseconds to delay boosting after RCU grace-period start"
213	range 0 3000
214	depends on RCU_BOOST
215	default 500
216	help
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.
221
222	  Accept the default if unsure.
223
224config 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
228	help
229	  Use this option to further reduce the latencies of expedited
230	  grace periods at the expense of being more disruptive.
231
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.
235
236	  Accept the default if unsure.
237
238config RCU_NOCB_CPU
239	bool "Offload RCU callback processing from boot-selected CPUs"
240	depends on TREE_RCU
241	depends on RCU_EXPERT || NO_HZ_FULL
242	default n
243	help
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
250	  rates.
251
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.
265
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.
270
271	  Say Y here if you need reduced OS jitter, despite added overhead.
272	  Say N here if you are unsure.
273
274config RCU_NOCB_CPU_DEFAULT_ALL
275	bool "Offload RCU callback processing from all CPUs by default"
276	depends on RCU_NOCB_CPU
277	default n
278	help
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.
283
284	  Say Y here if you want offload all CPUs by default on boot.
285	  Say N here if you are unsure.
286
287config 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
291	help
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.
299
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.
302
303config 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
307	help
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.
316
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.
320
321config RCU_LAZY
322	bool "RCU callback lazy invocation functionality"
323	depends on RCU_NOCB_CPU
324	default n
325	help
326	  To save power, batch RCU callbacks and flush after delay, memory
327	  pressure, or callback list growing too big.
328
329	  Requires rcu_nocbs=all to be set.
330
331	  Use rcutree.enable_rcu_lazy=0 to turn it off at boot time.
332
333config RCU_LAZY_DEFAULT_OFF
334	bool "Turn RCU lazy invocation off by default"
335	depends on RCU_LAZY
336	default n
337	help
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
340	  it back on.
341
342config RCU_DOUBLE_CHECK_CB_TIME
343	bool "RCU callback-batch backup time check"
344	depends on RCU_EXPERT
345	default n
346	help
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.
352
353	  This option rounds rcutree.rcu_resched_ns up to the next
354	  jiffy, and overrides the 32-callback batching if this limit
355	  is exceeded.
356
357	  Say Y here if you need tighter callback-limit enforcement.
358	  Say N here if you are unsure.
359
360endmenu # "RCU Subsystem"
361