projects/linux·blob v7.2-rc1 · refs

194 lines · 7.1 KiB

1# SPDX-License-Identifier: GPL-2.0-only
2
3config PREEMPT_NONE_BUILD
4 bool
5
6config PREEMPT_VOLUNTARY_BUILD
7 bool
8
9config PREEMPT_BUILD
10 bool
11 select PREEMPTION
12 select UNINLINE_SPIN_UNLOCK if !ARCH_INLINE_SPIN_UNLOCK
13
14config ARCH_HAS_PREEMPT_LAZY
15 bool
16
17choice
18 prompt "Preemption Model"
19 default PREEMPT_LAZY if ARCH_HAS_PREEMPT_LAZY
20 default PREEMPT_NONE
21
22config PREEMPT_NONE
23 bool "No Forced Preemption (Server)"
24 depends on !PREEMPT_RT
25 depends on ARCH_NO_PREEMPT
26 select PREEMPT_NONE_BUILD if !PREEMPT_DYNAMIC
27 help
28 This is the traditional Linux preemption model, geared towards
29 throughput. It will still provide good latencies most of the
30 time, but there are no guarantees and occasional longer delays
31 are possible.
32
33 Select this option if you are building a kernel for a server or
34 scientific/computation system, or if you want to maximize the
35 raw processing power of the kernel, irrespective of scheduling
36 latencies.
37
38config PREEMPT_VOLUNTARY
39 bool "Voluntary Kernel Preemption (Desktop)"
40 depends on !ARCH_HAS_PREEMPT_LAZY
41 depends on !ARCH_NO_PREEMPT
42 depends on !PREEMPT_RT
43 select PREEMPT_VOLUNTARY_BUILD if !PREEMPT_DYNAMIC
44 help
45 This option reduces the latency of the kernel by adding more
46 "explicit preemption points" to the kernel code. These new
47 preemption points have been selected to reduce the maximum
48 latency of rescheduling, providing faster application reactions,
49 at the cost of slightly lower throughput.
50
51 This allows reaction to interactive events by allowing a
52 low priority process to voluntarily preempt itself even if it
53 is in kernel mode executing a system call. This allows
54 applications to run more 'smoothly' even when the system is
55 under load.
56
57 Select this if you are building a kernel for a desktop system.
58
59config PREEMPT
60 bool "Preemptible Kernel (Low-Latency Desktop)"
61 depends on !ARCH_NO_PREEMPT
62 select PREEMPT_BUILD if !PREEMPT_DYNAMIC
63 help
64 This option reduces the latency of the kernel by making
65 all kernel code (that is not executing in a critical section)
66 preemptible. This allows reaction to interactive events by
67 permitting a low priority process to be preempted involuntarily
68 even if it is in kernel mode executing a system call and would
69 otherwise not be about to reach a natural preemption point.
70 This allows applications to run more 'smoothly' even when the
71 system is under load, at the cost of slightly lower throughput
72 and a slight runtime overhead to kernel code.
73
74 Select this if you are building a kernel for a desktop or
75 embedded system with latency requirements in the milliseconds
76 range.
77
78config PREEMPT_LAZY
79 bool "Scheduler controlled preemption model"
80 depends on !ARCH_NO_PREEMPT
81 depends on ARCH_HAS_PREEMPT_LAZY
82 select PREEMPT_BUILD if !PREEMPT_DYNAMIC
83 help
84 This option provides a scheduler driven preemption model that
85 is fundamentally similar to full preemption, but is less
86 eager to preempt SCHED_NORMAL tasks in an attempt to
87 reduce lock holder preemption and recover some of the performance
88 gains seen from using Voluntary preemption.
89
90endchoice
91
92config PREEMPT_RT
93 bool "Fully Preemptible Kernel (Real-Time)"
94 depends on EXPERT && ARCH_SUPPORTS_RT && !COMPILE_TEST
95 select PREEMPTION
96 help
97 This option turns the kernel into a real-time kernel by replacing
98 various locking primitives (spinlocks, rwlocks, etc.) with
99 preemptible priority-inheritance aware variants, enforcing
100 interrupt threading and introducing mechanisms to break up long
101 non-preemptible sections. This makes the kernel, except for very
102 low level and critical code paths (entry code, scheduler, low
103 level interrupt handling) fully preemptible and brings most
104 execution contexts under scheduler control.
105
106 Select this if you are building a kernel for systems which
107 require real-time guarantees.
108
109config PREEMPT_RT_NEEDS_BH_LOCK
110 bool "Enforce softirq synchronisation on PREEMPT_RT"
111 depends on PREEMPT_RT
112 help
113 Enforce synchronisation across the softirqs context. On PREEMPT_RT
114 the softirq is preemptible. This enforces the same per-CPU BLK
115 semantic non-PREEMPT_RT builds have. This should not be needed
116 because per-CPU locks were added to avoid the per-CPU BKL.
117
118 This switch provides the old behaviour for testing reasons. Select
119 this if you suspect an error with preemptible softirq and want test
120 the old synchronized behaviour.
121
122config PREEMPT_COUNT
123 bool
124
125config PREEMPTION
126 bool
127 select PREEMPT_COUNT
128
129config PREEMPT_DYNAMIC
130 bool "Preemption behaviour defined on boot"
131 depends on HAVE_PREEMPT_DYNAMIC
132 select JUMP_LABEL if HAVE_PREEMPT_DYNAMIC_KEY
133 select PREEMPT_BUILD
134 default y if HAVE_PREEMPT_DYNAMIC_CALL
135 help
136 This option allows to define the preemption model on the kernel
137 command line parameter and thus override the default preemption
138 model defined during compile time.
139
140 The feature is primarily interesting for Linux distributions which
141 provide a pre-built kernel binary to reduce the number of kernel
142 flavors they offer while still offering different usecases.
143
144 The runtime overhead is negligible with HAVE_STATIC_CALL_INLINE enabled
145 but if runtime patching is not available for the specific architecture
146 then the potential overhead should be considered.
147
148 Interesting if you want the same pre-built kernel should be used for
149 both Server and Desktop workloads.
150
151config SCHED_CORE
152 bool "Core Scheduling for SMT"
153 depends on SCHED_SMT
154 help
155 This option permits Core Scheduling, a means of coordinated task
156 selection across SMT siblings. When enabled -- see
157 prctl(PR_SCHED_CORE) -- task selection ensures that all SMT siblings
158 will execute a task from the same 'core group', forcing idle when no
159 matching task is found.
160
161 Use of this feature includes:
162 - mitigation of some (not all) SMT side channels;
163 - limiting SMT interference to improve determinism and/or performance.
164
165 SCHED_CORE is default disabled. When it is enabled and unused,
166 which is the likely usage by Linux distributions, there should
167 be no measurable impact on performance.
168
169config SCHED_CLASS_EXT
170 bool "Extensible Scheduling Class"
171 depends on BPF_SYSCALL && BPF_JIT && DEBUG_INFO_BTF
172 select STACKTRACE if STACKTRACE_SUPPORT
173 help
174 This option enables a new scheduler class sched_ext (SCX), which
175 allows scheduling policies to be implemented as BPF programs to
176 achieve the following:
177
178 - Ease of experimentation and exploration: Enabling rapid
179 iteration of new scheduling policies.
180 - Customization: Building application-specific schedulers which
181 implement policies that are not applicable to general-purpose
182 schedulers.
183 - Rapid scheduler deployments: Non-disruptive swap outs of
184 scheduling policies in production environments.
185
186 sched_ext leverages BPF struct_ops feature to define a structure
187 which exports function callbacks and flags to BPF programs that
188 wish to implement scheduling policies. The struct_ops structure
189 exported by sched_ext is struct sched_ext_ops, and is conceptually
190 similar to struct sched_class.
191
192 For more information:
193 Documentation/scheduler/sched-ext.rst
194 https://github.com/sched-ext/scx