0001 # SPDX-License-Identifier: GPL-2.0-only
0002
0003 config PREEMPT_NONE_BUILD
0004 bool
0005
0006 config PREEMPT_VOLUNTARY_BUILD
0007 bool
0008
0009 config PREEMPT_BUILD
0010 bool
0011 select PREEMPTION
0012 select UNINLINE_SPIN_UNLOCK if !ARCH_INLINE_SPIN_UNLOCK
0013
0014 choice
0015 prompt "Preemption Model"
0016 default PREEMPT_NONE
0017
0018 config PREEMPT_NONE
0019 bool "No Forced Preemption (Server)"
0020 select PREEMPT_NONE_BUILD if !PREEMPT_DYNAMIC
0021 help
0022 This is the traditional Linux preemption model, geared towards
0023 throughput. It will still provide good latencies most of the
0024 time, but there are no guarantees and occasional longer delays
0025 are possible.
0026
0027 Select this option if you are building a kernel for a server or
0028 scientific/computation system, or if you want to maximize the
0029 raw processing power of the kernel, irrespective of scheduling
0030 latencies.
0031
0032 config PREEMPT_VOLUNTARY
0033 bool "Voluntary Kernel Preemption (Desktop)"
0034 depends on !ARCH_NO_PREEMPT
0035 select PREEMPT_VOLUNTARY_BUILD if !PREEMPT_DYNAMIC
0036 help
0037 This option reduces the latency of the kernel by adding more
0038 "explicit preemption points" to the kernel code. These new
0039 preemption points have been selected to reduce the maximum
0040 latency of rescheduling, providing faster application reactions,
0041 at the cost of slightly lower throughput.
0042
0043 This allows reaction to interactive events by allowing a
0044 low priority process to voluntarily preempt itself even if it
0045 is in kernel mode executing a system call. This allows
0046 applications to run more 'smoothly' even when the system is
0047 under load.
0048
0049 Select this if you are building a kernel for a desktop system.
0050
0051 config PREEMPT
0052 bool "Preemptible Kernel (Low-Latency Desktop)"
0053 depends on !ARCH_NO_PREEMPT
0054 select PREEMPT_BUILD
0055 help
0056 This option reduces the latency of the kernel by making
0057 all kernel code (that is not executing in a critical section)
0058 preemptible. This allows reaction to interactive events by
0059 permitting a low priority process to be preempted involuntarily
0060 even if it is in kernel mode executing a system call and would
0061 otherwise not be about to reach a natural preemption point.
0062 This allows applications to run more 'smoothly' even when the
0063 system is under load, at the cost of slightly lower throughput
0064 and a slight runtime overhead to kernel code.
0065
0066 Select this if you are building a kernel for a desktop or
0067 embedded system with latency requirements in the milliseconds
0068 range.
0069
0070 config PREEMPT_RT
0071 bool "Fully Preemptible Kernel (Real-Time)"
0072 depends on EXPERT && ARCH_SUPPORTS_RT
0073 select PREEMPTION
0074 help
0075 This option turns the kernel into a real-time kernel by replacing
0076 various locking primitives (spinlocks, rwlocks, etc.) with
0077 preemptible priority-inheritance aware variants, enforcing
0078 interrupt threading and introducing mechanisms to break up long
0079 non-preemptible sections. This makes the kernel, except for very
0080 low level and critical code paths (entry code, scheduler, low
0081 level interrupt handling) fully preemptible and brings most
0082 execution contexts under scheduler control.
0083
0084 Select this if you are building a kernel for systems which
0085 require real-time guarantees.
0086
0087 endchoice
0088
0089 config PREEMPT_COUNT
0090 bool
0091
0092 config PREEMPTION
0093 bool
0094 select PREEMPT_COUNT
0095
0096 config PREEMPT_DYNAMIC
0097 bool "Preemption behaviour defined on boot"
0098 depends on HAVE_PREEMPT_DYNAMIC && !PREEMPT_RT
0099 select JUMP_LABEL if HAVE_PREEMPT_DYNAMIC_KEY
0100 select PREEMPT_BUILD
0101 default y if HAVE_PREEMPT_DYNAMIC_CALL
0102 help
0103 This option allows to define the preemption model on the kernel
0104 command line parameter and thus override the default preemption
0105 model defined during compile time.
0106
0107 The feature is primarily interesting for Linux distributions which
0108 provide a pre-built kernel binary to reduce the number of kernel
0109 flavors they offer while still offering different usecases.
0110
0111 The runtime overhead is negligible with HAVE_STATIC_CALL_INLINE enabled
0112 but if runtime patching is not available for the specific architecture
0113 then the potential overhead should be considered.
0114
0115 Interesting if you want the same pre-built kernel should be used for
0116 both Server and Desktop workloads.
0117
0118 config SCHED_CORE
0119 bool "Core Scheduling for SMT"
0120 depends on SCHED_SMT
0121 help
0122 This option permits Core Scheduling, a means of coordinated task
0123 selection across SMT siblings. When enabled -- see
0124 prctl(PR_SCHED_CORE) -- task selection ensures that all SMT siblings
0125 will execute a task from the same 'core group', forcing idle when no
0126 matching task is found.
0127
0128 Use of this feature includes:
0129 - mitigation of some (not all) SMT side channels;
0130 - limiting SMT interference to improve determinism and/or performance.
0131
0132 SCHED_CORE is default disabled. When it is enabled and unused,
0133 which is the likely usage by Linux distributions, there should
0134 be no measurable impact on performance.
0135
0136
