1 .. _sched_design_CFS: <<
2 <<
3 ============= 1 =============
4 CFS Scheduler 2 CFS Scheduler
5 ============= 3 =============
6 4
7 5
8 1. OVERVIEW 6 1. OVERVIEW
9 ============ 7 ============
10 8
11 CFS stands for "Completely Fair Scheduler," an !! 9 CFS stands for "Completely Fair Scheduler," and is the new "desktop" process
12 scheduler implemented by Ingo Molnar and merge !! 10 scheduler implemented by Ingo Molnar and merged in Linux 2.6.23. It is the
13 originally merged, it was the replacement for !! 11 replacement for the previous vanilla scheduler's SCHED_OTHER interactivity
14 scheduler's SCHED_OTHER interactivity code. No !! 12 code.
15 for EEVDF, for which documentation can be foun <<
16 Documentation/scheduler/sched-eevdf.rst. <<
17 13
18 80% of CFS's design can be summed up in a sing 14 80% of CFS's design can be summed up in a single sentence: CFS basically models
19 an "ideal, precise multi-tasking CPU" on real 15 an "ideal, precise multi-tasking CPU" on real hardware.
20 16
21 "Ideal multi-tasking CPU" is a (non-existent 17 "Ideal multi-tasking CPU" is a (non-existent :-)) CPU that has 100% physical
22 power and which can run each task at precise e 18 power and which can run each task at precise equal speed, in parallel, each at
23 1/nr_running speed. For example: if there are 19 1/nr_running speed. For example: if there are 2 tasks running, then it runs
24 each at 50% physical power --- i.e., actually 20 each at 50% physical power --- i.e., actually in parallel.
25 21
26 On real hardware, we can run only a single tas 22 On real hardware, we can run only a single task at once, so we have to
27 introduce the concept of "virtual runtime." T 23 introduce the concept of "virtual runtime." The virtual runtime of a task
28 specifies when its next timeslice would start 24 specifies when its next timeslice would start execution on the ideal
29 multi-tasking CPU described above. In practic 25 multi-tasking CPU described above. In practice, the virtual runtime of a task
30 is its actual runtime normalized to the total 26 is its actual runtime normalized to the total number of running tasks.
31 27
32 28
33 29
34 2. FEW IMPLEMENTATION DETAILS 30 2. FEW IMPLEMENTATION DETAILS
35 ============================== 31 ==============================
36 32
37 In CFS the virtual runtime is expressed and tr 33 In CFS the virtual runtime is expressed and tracked via the per-task
38 p->se.vruntime (nanosec-unit) value. This way 34 p->se.vruntime (nanosec-unit) value. This way, it's possible to accurately
39 timestamp and measure the "expected CPU time" 35 timestamp and measure the "expected CPU time" a task should have gotten.
40 36
41 Small detail: on "ideal" hardware, at any t 37 Small detail: on "ideal" hardware, at any time all tasks would have the same
42 p->se.vruntime value --- i.e., tasks would 38 p->se.vruntime value --- i.e., tasks would execute simultaneously and no task
43 would ever get "out of balance" from the "i 39 would ever get "out of balance" from the "ideal" share of CPU time.
44 40
45 CFS's task picking logic is based on this p->s 41 CFS's task picking logic is based on this p->se.vruntime value and it is thus
46 very simple: it always tries to run the task w 42 very simple: it always tries to run the task with the smallest p->se.vruntime
47 value (i.e., the task which executed least so 43 value (i.e., the task which executed least so far). CFS always tries to split
48 up CPU time between runnable tasks as close to 44 up CPU time between runnable tasks as close to "ideal multitasking hardware" as
49 possible. 45 possible.
50 46
51 Most of the rest of CFS's design just falls ou 47 Most of the rest of CFS's design just falls out of this really simple concept,
52 with a few add-on embellishments like nice lev 48 with a few add-on embellishments like nice levels, multiprocessing and various
53 algorithm variants to recognize sleepers. 49 algorithm variants to recognize sleepers.
54 50
55 51
56 52
57 3. THE RBTREE 53 3. THE RBTREE
58 ============== 54 ==============
59 55
60 CFS's design is quite radical: it does not use 56 CFS's design is quite radical: it does not use the old data structures for the
61 runqueues, but it uses a time-ordered rbtree t 57 runqueues, but it uses a time-ordered rbtree to build a "timeline" of future
62 task execution, and thus has no "array switch" 58 task execution, and thus has no "array switch" artifacts (by which both the
63 previous vanilla scheduler and RSDL/SD are aff 59 previous vanilla scheduler and RSDL/SD are affected).
64 60
65 CFS also maintains the rq->cfs.min_vruntime va 61 CFS also maintains the rq->cfs.min_vruntime value, which is a monotonic
66 increasing value tracking the smallest vruntim 62 increasing value tracking the smallest vruntime among all tasks in the
67 runqueue. The total amount of work done by th 63 runqueue. The total amount of work done by the system is tracked using
68 min_vruntime; that value is used to place newl 64 min_vruntime; that value is used to place newly activated entities on the left
69 side of the tree as much as possible. 65 side of the tree as much as possible.
70 66
71 The total number of running tasks in the runqu 67 The total number of running tasks in the runqueue is accounted through the
72 rq->cfs.load value, which is the sum of the we 68 rq->cfs.load value, which is the sum of the weights of the tasks queued on the
73 runqueue. 69 runqueue.
74 70
75 CFS maintains a time-ordered rbtree, where all 71 CFS maintains a time-ordered rbtree, where all runnable tasks are sorted by the
76 p->se.vruntime key. CFS picks the "leftmost" t 72 p->se.vruntime key. CFS picks the "leftmost" task from this tree and sticks to it.
77 As the system progresses forwards, the execute 73 As the system progresses forwards, the executed tasks are put into the tree
78 more and more to the right --- slowly but sure 74 more and more to the right --- slowly but surely giving a chance for every task
79 to become the "leftmost task" and thus get on 75 to become the "leftmost task" and thus get on the CPU within a deterministic
80 amount of time. 76 amount of time.
81 77
82 Summing up, CFS works like this: it runs a tas 78 Summing up, CFS works like this: it runs a task a bit, and when the task
83 schedules (or a scheduler tick happens) the ta 79 schedules (or a scheduler tick happens) the task's CPU usage is "accounted
84 for": the (small) time it just spent using the 80 for": the (small) time it just spent using the physical CPU is added to
85 p->se.vruntime. Once p->se.vruntime gets high 81 p->se.vruntime. Once p->se.vruntime gets high enough so that another task
86 becomes the "leftmost task" of the time-ordere 82 becomes the "leftmost task" of the time-ordered rbtree it maintains (plus a
87 small amount of "granularity" distance relativ 83 small amount of "granularity" distance relative to the leftmost task so that we
88 do not over-schedule tasks and trash the cache 84 do not over-schedule tasks and trash the cache), then the new leftmost task is
89 picked and the current task is preempted. 85 picked and the current task is preempted.
90 86
91 87
92 88
93 4. SOME FEATURES OF CFS 89 4. SOME FEATURES OF CFS
94 ======================== 90 ========================
95 91
96 CFS uses nanosecond granularity accounting and 92 CFS uses nanosecond granularity accounting and does not rely on any jiffies or
97 other HZ detail. Thus the CFS scheduler has n 93 other HZ detail. Thus the CFS scheduler has no notion of "timeslices" in the
98 way the previous scheduler had, and has no heu 94 way the previous scheduler had, and has no heuristics whatsoever. There is
99 only one central tunable (you have to switch o 95 only one central tunable (you have to switch on CONFIG_SCHED_DEBUG):
100 96
101 /sys/kernel/debug/sched/base_slice_ns 97 /sys/kernel/debug/sched/base_slice_ns
102 98
103 which can be used to tune the scheduler from " 99 which can be used to tune the scheduler from "desktop" (i.e., low latencies) to
104 "server" (i.e., good batching) workloads. It 100 "server" (i.e., good batching) workloads. It defaults to a setting suitable
105 for desktop workloads. SCHED_BATCH is handled 101 for desktop workloads. SCHED_BATCH is handled by the CFS scheduler module too.
106 102
107 In case CONFIG_HZ results in base_slice_ns < T 103 In case CONFIG_HZ results in base_slice_ns < TICK_NSEC, the value of
108 base_slice_ns will have little to no impact on 104 base_slice_ns will have little to no impact on the workloads.
109 105
110 Due to its design, the CFS scheduler is not pr 106 Due to its design, the CFS scheduler is not prone to any of the "attacks" that
111 exist today against the heuristics of the stoc 107 exist today against the heuristics of the stock scheduler: fiftyp.c, thud.c,
112 chew.c, ring-test.c, massive_intr.c all work f 108 chew.c, ring-test.c, massive_intr.c all work fine and do not impact
113 interactivity and produce the expected behavio 109 interactivity and produce the expected behavior.
114 110
115 The CFS scheduler has a much stronger handling 111 The CFS scheduler has a much stronger handling of nice levels and SCHED_BATCH
116 than the previous vanilla scheduler: both type 112 than the previous vanilla scheduler: both types of workloads are isolated much
117 more aggressively. 113 more aggressively.
118 114
119 SMP load-balancing has been reworked/sanitized 115 SMP load-balancing has been reworked/sanitized: the runqueue-walking
120 assumptions are gone from the load-balancing c 116 assumptions are gone from the load-balancing code now, and iterators of the
121 scheduling modules are used. The balancing co 117 scheduling modules are used. The balancing code got quite a bit simpler as a
122 result. 118 result.
123 119
124 120
125 121
126 5. Scheduling policies 122 5. Scheduling policies
127 ====================== 123 ======================
128 124
129 CFS implements three scheduling policies: 125 CFS implements three scheduling policies:
130 126
131 - SCHED_NORMAL (traditionally called SCHED_O 127 - SCHED_NORMAL (traditionally called SCHED_OTHER): The scheduling
132 policy that is used for regular tasks. 128 policy that is used for regular tasks.
133 129
134 - SCHED_BATCH: Does not preempt nearly as of 130 - SCHED_BATCH: Does not preempt nearly as often as regular tasks
135 would, thereby allowing tasks to run longe 131 would, thereby allowing tasks to run longer and make better use of
136 caches but at the cost of interactivity. T 132 caches but at the cost of interactivity. This is well suited for
137 batch jobs. 133 batch jobs.
138 134
139 - SCHED_IDLE: This is even weaker than nice 135 - SCHED_IDLE: This is even weaker than nice 19, but its not a true
140 idle timer scheduler in order to avoid to 136 idle timer scheduler in order to avoid to get into priority
141 inversion problems which would deadlock th 137 inversion problems which would deadlock the machine.
142 138
143 SCHED_FIFO/_RR are implemented in sched/rt.c a 139 SCHED_FIFO/_RR are implemented in sched/rt.c and are as specified by
144 POSIX. 140 POSIX.
145 141
146 The command chrt from util-linux-ng 2.13.1.1 c 142 The command chrt from util-linux-ng 2.13.1.1 can set all of these except
147 SCHED_IDLE. 143 SCHED_IDLE.
148 144
149 145
150 146
151 6. SCHEDULING CLASSES 147 6. SCHEDULING CLASSES
152 ====================== 148 ======================
153 149
154 The new CFS scheduler has been designed in suc 150 The new CFS scheduler has been designed in such a way to introduce "Scheduling
155 Classes," an extensible hierarchy of scheduler 151 Classes," an extensible hierarchy of scheduler modules. These modules
156 encapsulate scheduling policy details and are 152 encapsulate scheduling policy details and are handled by the scheduler core
157 without the core code assuming too much about 153 without the core code assuming too much about them.
158 154
159 sched/fair.c implements the CFS scheduler desc 155 sched/fair.c implements the CFS scheduler described above.
160 156
161 sched/rt.c implements SCHED_FIFO and SCHED_RR 157 sched/rt.c implements SCHED_FIFO and SCHED_RR semantics, in a simpler way than
162 the previous vanilla scheduler did. It uses 1 158 the previous vanilla scheduler did. It uses 100 runqueues (for all 100 RT
163 priority levels, instead of 140 in the previou 159 priority levels, instead of 140 in the previous scheduler) and it needs no
164 expired array. 160 expired array.
165 161
166 Scheduling classes are implemented through the 162 Scheduling classes are implemented through the sched_class structure, which
167 contains hooks to functions that must be calle 163 contains hooks to functions that must be called whenever an interesting event
168 occurs. 164 occurs.
169 165
170 This is the (partial) list of the hooks: 166 This is the (partial) list of the hooks:
171 167
172 - enqueue_task(...) 168 - enqueue_task(...)
173 169
174 Called when a task enters a runnable state. 170 Called when a task enters a runnable state.
175 It puts the scheduling entity (task) into t 171 It puts the scheduling entity (task) into the red-black tree and
176 increments the nr_running variable. 172 increments the nr_running variable.
177 173
178 - dequeue_task(...) 174 - dequeue_task(...)
179 175
180 When a task is no longer runnable, this fun 176 When a task is no longer runnable, this function is called to keep the
181 corresponding scheduling entity out of the 177 corresponding scheduling entity out of the red-black tree. It decrements
182 the nr_running variable. 178 the nr_running variable.
183 179
184 - yield_task(...) 180 - yield_task(...)
185 181
186 This function is basically just a dequeue f 182 This function is basically just a dequeue followed by an enqueue, unless the
187 compat_yield sysctl is turned on; in that c 183 compat_yield sysctl is turned on; in that case, it places the scheduling
188 entity at the right-most end of the red-bla 184 entity at the right-most end of the red-black tree.
189 185
190 - wakeup_preempt(...) 186 - wakeup_preempt(...)
191 187
192 This function checks if a task that entered 188 This function checks if a task that entered the runnable state should
193 preempt the currently running task. 189 preempt the currently running task.
194 190
195 - pick_next_task(...) 191 - pick_next_task(...)
196 192
197 This function chooses the most appropriate 193 This function chooses the most appropriate task eligible to run next.
198 194
199 - set_next_task(...) 195 - set_next_task(...)
200 196
201 This function is called when a task changes 197 This function is called when a task changes its scheduling class, changes
202 its task group or is scheduled. 198 its task group or is scheduled.
203 199
204 - task_tick(...) 200 - task_tick(...)
205 201
206 This function is mostly called from time ti 202 This function is mostly called from time tick functions; it might lead to
207 process switch. This drives the running pr 203 process switch. This drives the running preemption.
208 204
209 205
210 206
211 207
212 7. GROUP SCHEDULER EXTENSIONS TO CFS 208 7. GROUP SCHEDULER EXTENSIONS TO CFS
213 ===================================== 209 =====================================
214 210
215 Normally, the scheduler operates on individual 211 Normally, the scheduler operates on individual tasks and strives to provide
216 fair CPU time to each task. Sometimes, it may 212 fair CPU time to each task. Sometimes, it may be desirable to group tasks and
217 provide fair CPU time to each such task group. 213 provide fair CPU time to each such task group. For example, it may be
218 desirable to first provide fair CPU time to ea 214 desirable to first provide fair CPU time to each user on the system and then to
219 each task belonging to a user. 215 each task belonging to a user.
220 216
221 CONFIG_CGROUP_SCHED strives to achieve exactly 217 CONFIG_CGROUP_SCHED strives to achieve exactly that. It lets tasks to be
222 grouped and divides CPU time fairly among such 218 grouped and divides CPU time fairly among such groups.
223 219
224 CONFIG_RT_GROUP_SCHED permits to group real-ti 220 CONFIG_RT_GROUP_SCHED permits to group real-time (i.e., SCHED_FIFO and
225 SCHED_RR) tasks. 221 SCHED_RR) tasks.
226 222
227 CONFIG_FAIR_GROUP_SCHED permits to group CFS ( 223 CONFIG_FAIR_GROUP_SCHED permits to group CFS (i.e., SCHED_NORMAL and
228 SCHED_BATCH) tasks. 224 SCHED_BATCH) tasks.
229 225
230 These options need CONFIG_CGROUPS to be def 226 These options need CONFIG_CGROUPS to be defined, and let the administrator
231 create arbitrary groups of tasks, using the 227 create arbitrary groups of tasks, using the "cgroup" pseudo filesystem. See
232 Documentation/admin-guide/cgroup-v1/cgroups 228 Documentation/admin-guide/cgroup-v1/cgroups.rst for more information about this filesystem.
233 229
234 When CONFIG_FAIR_GROUP_SCHED is defined, a "cp 230 When CONFIG_FAIR_GROUP_SCHED is defined, a "cpu.shares" file is created for each
235 group created using the pseudo filesystem. Se 231 group created using the pseudo filesystem. See example steps below to create
236 task groups and modify their CPU share using t 232 task groups and modify their CPU share using the "cgroups" pseudo filesystem::
237 233
238 # mount -t tmpfs cgroup_root /sys/fs/c 234 # mount -t tmpfs cgroup_root /sys/fs/cgroup
239 # mkdir /sys/fs/cgroup/cpu 235 # mkdir /sys/fs/cgroup/cpu
240 # mount -t cgroup -ocpu none /sys/fs/c 236 # mount -t cgroup -ocpu none /sys/fs/cgroup/cpu
241 # cd /sys/fs/cgroup/cpu 237 # cd /sys/fs/cgroup/cpu
242 238
243 # mkdir multimedia # create "mult 239 # mkdir multimedia # create "multimedia" group of tasks
244 # mkdir browser # create "brow 240 # mkdir browser # create "browser" group of tasks
245 241
246 # #Configure the multimedia group to r 242 # #Configure the multimedia group to receive twice the CPU bandwidth
247 # #that of browser group 243 # #that of browser group
248 244
249 # echo 2048 > multimedia/cpu.shares 245 # echo 2048 > multimedia/cpu.shares
250 # echo 1024 > browser/cpu.shares 246 # echo 1024 > browser/cpu.shares
251 247
252 # firefox & # Launch firefox and m 248 # firefox & # Launch firefox and move it to "browser" group
253 # echo <firefox_pid> > browser/tasks 249 # echo <firefox_pid> > browser/tasks
254 250
255 # #Launch gmplayer (or your favourite 251 # #Launch gmplayer (or your favourite movie player)
256 # echo <movie_player_pid> > multimedia 252 # echo <movie_player_pid> > multimedia/tasks
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.