projects/linux·blob v6.19 · refs

191 lines · 7.0 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_NONE
20
21config PREEMPT_NONE
22 bool "No Forced Preemption (Server)"
23 depends on !PREEMPT_RT
24 select PREEMPT_NONE_BUILD if !PREEMPT_DYNAMIC
25 help
26 This is the traditional Linux preemption model, geared towards
27 throughput. It will still provide good latencies most of the
28 time, but there are no guarantees and occasional longer delays
29 are possible.
30
31 Select this option if you are building a kernel for a server or
32 scientific/computation system, or if you want to maximize the
33 raw processing power of the kernel, irrespective of scheduling
34 latencies.
35
36config PREEMPT_VOLUNTARY
37 bool "Voluntary Kernel Preemption (Desktop)"
38 depends on !ARCH_NO_PREEMPT
39 depends on !PREEMPT_RT
40 select PREEMPT_VOLUNTARY_BUILD if !PREEMPT_DYNAMIC
41 help
42 This option reduces the latency of the kernel by adding more
43 "explicit preemption points" to the kernel code. These new
44 preemption points have been selected to reduce the maximum
45 latency of rescheduling, providing faster application reactions,
46 at the cost of slightly lower throughput.
47
48 This allows reaction to interactive events by allowing a
49 low priority process to voluntarily preempt itself even if it
50 is in kernel mode executing a system call. This allows
51 applications to run more 'smoothly' even when the system is
52 under load.
53
54 Select this if you are building a kernel for a desktop system.
55
56config PREEMPT
57 bool "Preemptible Kernel (Low-Latency Desktop)"
58 depends on !ARCH_NO_PREEMPT
59 select PREEMPT_BUILD if !PREEMPT_DYNAMIC
60 help
61 This option reduces the latency of the kernel by making
62 all kernel code (that is not executing in a critical section)
63 preemptible. This allows reaction to interactive events by
64 permitting a low priority process to be preempted involuntarily
65 even if it is in kernel mode executing a system call and would
66 otherwise not be about to reach a natural preemption point.
67 This allows applications to run more 'smoothly' even when the
68 system is under load, at the cost of slightly lower throughput
69 and a slight runtime overhead to kernel code.
70
71 Select this if you are building a kernel for a desktop or
72 embedded system with latency requirements in the milliseconds
73 range.
74
75config PREEMPT_LAZY
76 bool "Scheduler controlled preemption model"
77 depends on !ARCH_NO_PREEMPT
78 depends on ARCH_HAS_PREEMPT_LAZY
79 select PREEMPT_BUILD if !PREEMPT_DYNAMIC
80 help
81 This option provides a scheduler driven preemption model that
82 is fundamentally similar to full preemption, but is less
83 eager to preempt SCHED_NORMAL tasks in an attempt to
84 reduce lock holder preemption and recover some of the performance
85 gains seen from using Voluntary preemption.
86
87endchoice
88
89config PREEMPT_RT
90 bool "Fully Preemptible Kernel (Real-Time)"
91 depends on EXPERT && ARCH_SUPPORTS_RT && !COMPILE_TEST
92 select PREEMPTION
93 help
94 This option turns the kernel into a real-time kernel by replacing
95 various locking primitives (spinlocks, rwlocks, etc.) with
96 preemptible priority-inheritance aware variants, enforcing
97 interrupt threading and introducing mechanisms to break up long
98 non-preemptible sections. This makes the kernel, except for very
99 low level and critical code paths (entry code, scheduler, low
100 level interrupt handling) fully preemptible and brings most
101 execution contexts under scheduler control.
102
103 Select this if you are building a kernel for systems which
104 require real-time guarantees.
105
106config PREEMPT_RT_NEEDS_BH_LOCK
107 bool "Enforce softirq synchronisation on PREEMPT_RT"
108 depends on PREEMPT_RT
109 help
110 Enforce synchronisation across the softirqs context. On PREEMPT_RT
111 the softirq is preemptible. This enforces the same per-CPU BLK
112 semantic non-PREEMPT_RT builds have. This should not be needed
113 because per-CPU locks were added to avoid the per-CPU BKL.
114
115 This switch provides the old behaviour for testing reasons. Select
116 this if you suspect an error with preemptible softirq and want test
117 the old synchronized behaviour.
118
119config PREEMPT_COUNT
120 bool
121
122config PREEMPTION
123 bool
124 select PREEMPT_COUNT
125
126config PREEMPT_DYNAMIC
127 bool "Preemption behaviour defined on boot"
128 depends on HAVE_PREEMPT_DYNAMIC
129 select JUMP_LABEL if HAVE_PREEMPT_DYNAMIC_KEY
130 select PREEMPT_BUILD
131 default y if HAVE_PREEMPT_DYNAMIC_CALL
132 help
133 This option allows to define the preemption model on the kernel
134 command line parameter and thus override the default preemption
135 model defined during compile time.
136
137 The feature is primarily interesting for Linux distributions which
138 provide a pre-built kernel binary to reduce the number of kernel
139 flavors they offer while still offering different usecases.
140
141 The runtime overhead is negligible with HAVE_STATIC_CALL_INLINE enabled
142 but if runtime patching is not available for the specific architecture
143 then the potential overhead should be considered.
144
145 Interesting if you want the same pre-built kernel should be used for
146 both Server and Desktop workloads.
147
148config SCHED_CORE
149 bool "Core Scheduling for SMT"
150 depends on SCHED_SMT
151 help
152 This option permits Core Scheduling, a means of coordinated task
153 selection across SMT siblings. When enabled -- see
154 prctl(PR_SCHED_CORE) -- task selection ensures that all SMT siblings
155 will execute a task from the same 'core group', forcing idle when no
156 matching task is found.
157
158 Use of this feature includes:
159 - mitigation of some (not all) SMT side channels;
160 - limiting SMT interference to improve determinism and/or performance.
161
162 SCHED_CORE is default disabled. When it is enabled and unused,
163 which is the likely usage by Linux distributions, there should
164 be no measurable impact on performance.
165
166config SCHED_CLASS_EXT
167 bool "Extensible Scheduling Class"
168 depends on BPF_SYSCALL && BPF_JIT && DEBUG_INFO_BTF
169 select STACKTRACE if STACKTRACE_SUPPORT
170 help
171 This option enables a new scheduler class sched_ext (SCX), which
172 allows scheduling policies to be implemented as BPF programs to
173 achieve the following:
174
175 - Ease of experimentation and exploration: Enabling rapid
176 iteration of new scheduling policies.
177 - Customization: Building application-specific schedulers which
178 implement policies that are not applicable to general-purpose
179 schedulers.
180 - Rapid scheduler deployments: Non-disruptive swap outs of
181 scheduling policies in production environments.
182
183 sched_ext leverages BPF struct_ops feature to define a structure
184 which exports function callbacks and flags to BPF programs that
185 wish to implement scheduling policies. The struct_ops structure
186 exported by sched_ext is struct sched_ext_ops, and is conceptually
187 similar to struct sched_class.
188
189 For more information:
190 Documentation/scheduler/sched-ext.rst
191 https://github.com/sched-ext/scx