1 ========================
2 Deadline Task Scheduling
3 ========================
4
5 .. CONTENTS
6
7 0. WARNING
8 1. Overview
9 2. Scheduling algorithm
10 2.1 Main algorithm
11 2.2 Bandwidth reclaiming
12 3. Scheduling Real-Time Tasks
13 3.1 Definitions
14 3.2 Schedulability Analysis for Uniprocessor Systems
15 3.3 Schedulability Analysis for Multiprocessor Systems
16 3.4 Relationship with SCHED_DEADLINE Parameters
17 4. Bandwidth management
18 4.1 System-wide settings
19 4.2 Task interface
20 4.3 Default behavior
21 4.4 Behavior of sched_yield()
22 5. Tasks CPU affinity
23 5.1 SCHED_DEADLINE and cpusets HOWTO
24 6. Future plans
25 A. Test suite
26 B. Minimal main()
27
28
29 0. WARNING
30 ==========
31
32 Fiddling with these settings can result in an unpredictable or even unstable
33 system behavior. As for -rt (group) scheduling, it is assumed that root users
34 know what they're doing.
35
36
37 1. Overview
38 ===========
39
40 The SCHED_DEADLINE policy contained inside the sched_dl scheduling class is
41 basically an implementation of the Earliest Deadline First (EDF) scheduling
42 algorithm, augmented with a mechanism (called Constant Bandwidth Server, CBS)
43 that makes it possible to isolate the behavior of tasks between each other.
44
45
46 2. Scheduling algorithm
47 =======================
48
49 2.1 Main algorithm
50 ------------------
51
52 SCHED_DEADLINE [18] uses three parameters, named "runtime", "period", and
53 "deadline", to schedule tasks. A SCHED_DEADLINE task should receive
54 "runtime" microseconds of execution time every "period" microseconds, and
55 these "runtime" microseconds are available within "deadline" microseconds
56 from the beginning of the period. In order to implement this behavior,
57 every time the task wakes up, the scheduler computes a "scheduling deadline"
58 consistent with the guarantee (using the CBS[2,3] algorithm). Tasks are then
59 scheduled using EDF[1] on these scheduling deadlines (the task with the
60 earliest scheduling deadline is selected for execution). Notice that the
61 task actually receives "runtime" time units within "deadline" if a proper
62 "admission control" strategy (see Section "4. Bandwidth management") is used
63 (clearly, if the system is overloaded this guarantee cannot be respected).
64
65 Summing up, the CBS[2,3] algorithm assigns scheduling deadlines to tasks so
66 that each task runs for at most its runtime every period, avoiding any
67 interference between different tasks (bandwidth isolation), while the EDF[1]
68 algorithm selects the task with the earliest scheduling deadline as the one
69 to be executed next. Thanks to this feature, tasks that do not strictly comply
70 with the "traditional" real-time task model (see Section 3) can effectively
71 use the new policy.
72
73 In more details, the CBS algorithm assigns scheduling deadlines to
74 tasks in the following way:
75
76 - Each SCHED_DEADLINE task is characterized by the "runtime",
77 "deadline", and "period" parameters;
78
79 - The state of the task is described by a "scheduling deadline", and
80 a "remaining runtime". These two parameters are initially set to 0;
81
82 - When a SCHED_DEADLINE task wakes up (becomes ready for execution),
83 the scheduler checks if::
84
85 remaining runtime runtime
86 ---------------------------------- > ---------
87 scheduling deadline - current time period
88
89 then, if the scheduling deadline is smaller than the current time, or
90 this condition is verified, the scheduling deadline and the
91 remaining runtime are re-initialized as
92
93 scheduling deadline = current time + deadline
94 remaining runtime = runtime
95
96 otherwise, the scheduling deadline and the remaining runtime are
97 left unchanged;
98
99 - When a SCHED_DEADLINE task executes for an amount of time t, its
100 remaining runtime is decreased as::
101
102 remaining runtime = remaining runtime - t
103
104 (technically, the runtime is decreased at every tick, or when the
105 task is descheduled / preempted);
106
107 - When the remaining runtime becomes less or equal than 0, the task is
108 said to be "throttled" (also known as "depleted" in real-time literature)
109 and cannot be scheduled until its scheduling deadline. The "replenishment
110 time" for this task (see next item) is set to be equal to the current
111 value of the scheduling deadline;
112
113 - When the current time is equal to the replenishment time of a
114 throttled task, the scheduling deadline and the remaining runtime are
115 updated as::
116
117 scheduling deadline = scheduling deadline + period
118 remaining runtime = remaining runtime + runtime
119
120 The SCHED_FLAG_DL_OVERRUN flag in sched_attr's sched_flags field allows a task
121 to get informed about runtime overruns through the delivery of SIGXCPU
122 signals.
123
124
125 2.2 Bandwidth reclaiming
126 ------------------------
127
128 Bandwidth reclaiming for deadline tasks is based on the GRUB (Greedy
129 Reclamation of Unused Bandwidth) algorithm [15, 16, 17] and it is enabled
130 when flag SCHED_FLAG_RECLAIM is set.
131
132 The following diagram illustrates the state names for tasks handled by GRUB::
133
134 ------------
135 (d) | Active |
136 ------------->| |
137 | | Contending |
138 | ------------
139 | A |
140 ---------- | |
141 | | | |
142 | Inactive | |(b) | (a)
143 | | | |
144 ---------- | |
145 A | V
146 | ------------
147 | | Active |
148 --------------| Non |
149 (c) | Contending |
150 ------------
151
152 A task can be in one of the following states:
153
154 - ActiveContending: if it is ready for execution (or executing);
155
156 - ActiveNonContending: if it just blocked and has not yet surpassed the 0-lag
157 time;
158
159 - Inactive: if it is blocked and has surpassed the 0-lag time.
160
161 State transitions:
162
163 (a) When a task blocks, it does not become immediately inactive since its
164 bandwidth cannot be immediately reclaimed without breaking the
165 real-time guarantees. It therefore enters a transitional state called
166 ActiveNonContending. The scheduler arms the "inactive timer" to fire at
167 the 0-lag time, when the task's bandwidth can be reclaimed without
168 breaking the real-time guarantees.
169
170 The 0-lag time for a task entering the ActiveNonContending state is
171 computed as::
172
173 (runtime * dl_period)
174 deadline - ---------------------
175 dl_runtime
176
177 where runtime is the remaining runtime, while dl_runtime and dl_period
178 are the reservation parameters.
179
180 (b) If the task wakes up before the inactive timer fires, the task re-enters
181 the ActiveContending state and the "inactive timer" is canceled.
182 In addition, if the task wakes up on a different runqueue, then
183 the task's utilization must be removed from the previous runqueue's active
184 utilization and must be added to the new runqueue's active utilization.
185 In order to avoid races between a task waking up on a runqueue while the
186 "inactive timer" is running on a different CPU, the "dl_non_contending"
187 flag is used to indicate that a task is not on a runqueue but is active
188 (so, the flag is set when the task blocks and is cleared when the
189 "inactive timer" fires or when the task wakes up).
190
191 (c) When the "inactive timer" fires, the task enters the Inactive state and
192 its utilization is removed from the runqueue's active utilization.
193
194 (d) When an inactive task wakes up, it enters the ActiveContending state and
195 its utilization is added to the active utilization of the runqueue where
196 it has been enqueued.
197
198 For each runqueue, the algorithm GRUB keeps track of two different bandwidths:
199
200 - Active bandwidth (running_bw): this is the sum of the bandwidths of all
201 tasks in active state (i.e., ActiveContending or ActiveNonContending);
202
203 - Total bandwidth (this_bw): this is the sum of all tasks "belonging" to the
204 runqueue, including the tasks in Inactive state.
205
206 - Maximum usable bandwidth (max_bw): This is the maximum bandwidth usable by
207 deadline tasks and is currently set to the RT capacity.
208
209
210 The algorithm reclaims the bandwidth of the tasks in Inactive state.
211 It does so by decrementing the runtime of the executing task Ti at a pace equal
212 to
213
214 dq = -(max{ Ui, (Umax - Uinact - Uextra) } / Umax) dt
215
216 where:
217
218 - Ui is the bandwidth of task Ti;
219 - Umax is the maximum reclaimable utilization (subjected to RT throttling
220 limits);
221 - Uinact is the (per runqueue) inactive utilization, computed as
222 (this_bq - running_bw);
223 - Uextra is the (per runqueue) extra reclaimable utilization
224 (subjected to RT throttling limits).
225
226
227 Let's now see a trivial example of two deadline tasks with runtime equal
228 to 4 and period equal to 8 (i.e., bandwidth equal to 0.5)::
229
230 A Task T1
231 |
232 | |
233 | |
234 |-------- |----
235 | | V
236 |---|---|---|---|---|---|---|---|--------->t
237 0 1 2 3 4 5 6 7 8
238
239
240 A Task T2
241 |
242 | |
243 | |
244 | ------------------------|
245 | | V
246 |---|---|---|---|---|---|---|---|--------->t
247 0 1 2 3 4 5 6 7 8
248
249
250 A running_bw
251 |
252 1 ----------------- ------
253 | | |
254 0.5- -----------------
255 | |
256 |---|---|---|---|---|---|---|---|--------->t
257 0 1 2 3 4 5 6 7 8
258
259
260 - Time t = 0:
261
262 Both tasks are ready for execution and therefore in ActiveContending state.
263 Suppose Task T1 is the first task to start execution.
264 Since there are no inactive tasks, its runtime is decreased as dq = -1 dt.
265
266 - Time t = 2:
267
268 Suppose that task T1 blocks
269 Task T1 therefore enters the ActiveNonContending state. Since its remaining
270 runtime is equal to 2, its 0-lag time is equal to t = 4.
271 Task T2 start execution, with runtime still decreased as dq = -1 dt since
272 there are no inactive tasks.
273
274 - Time t = 4:
275
276 This is the 0-lag time for Task T1. Since it didn't woken up in the
277 meantime, it enters the Inactive state. Its bandwidth is removed from
278 running_bw.
279 Task T2 continues its execution. However, its runtime is now decreased as
280 dq = - 0.5 dt because Uinact = 0.5.
281 Task T2 therefore reclaims the bandwidth unused by Task T1.
282
283 - Time t = 8:
284
285 Task T1 wakes up. It enters the ActiveContending state again, and the
286 running_bw is incremented.
287
288
289 2.3 Energy-aware scheduling
290 ---------------------------
291
292 When cpufreq's schedutil governor is selected, SCHED_DEADLINE implements the
293 GRUB-PA [19] algorithm, reducing the CPU operating frequency to the minimum
294 value that still allows to meet the deadlines. This behavior is currently
295 implemented only for ARM architectures.
296
297 A particular care must be taken in case the time needed for changing frequency
298 is of the same order of magnitude of the reservation period. In such cases,
299 setting a fixed CPU frequency results in a lower amount of deadline misses.
300
301
302 3. Scheduling Real-Time Tasks
303 =============================
304
305
306
307 .. BIG FAT WARNING ******************************************************
308
309 .. warning::
310
311 This section contains a (not-thorough) summary on classical deadline
312 scheduling theory, and how it applies to SCHED_DEADLINE.
313 The reader can "safely" skip to Section 4 if only interested in seeing
314 how the scheduling policy can be used. Anyway, we strongly recommend
315 to come back here and continue reading (once the urge for testing is
316 satisfied :P) to be sure of fully understanding all technical details.
317
318 .. ************************************************************************
319
320 There are no limitations on what kind of task can exploit this new
321 scheduling discipline, even if it must be said that it is particularly
322 suited for periodic or sporadic real-time tasks that need guarantees on their
323 timing behavior, e.g., multimedia, streaming, control applications, etc.
324
325 3.1 Definitions
326 ------------------------
327
328 A typical real-time task is composed of a repetition of computation phases
329 (task instances, or jobs) which are activated on a periodic or sporadic
330 fashion.
331 Each job J_j (where J_j is the j^th job of the task) is characterized by an
332 arrival time r_j (the time when the job starts), an amount of computation
333 time c_j needed to finish the job, and a job absolute deadline d_j, which
334 is the time within which the job should be finished. The maximum execution
335 time max{c_j} is called "Worst Case Execution Time" (WCET) for the task.
336 A real-time task can be periodic with period P if r_{j+1} = r_j + P, or
337 sporadic with minimum inter-arrival time P is r_{j+1} >= r_j + P. Finally,
338 d_j = r_j + D, where D is the task's relative deadline.
339 Summing up, a real-time task can be described as
340
341 Task = (WCET, D, P)
342
343 The utilization of a real-time task is defined as the ratio between its
344 WCET and its period (or minimum inter-arrival time), and represents
345 the fraction of CPU time needed to execute the task.
346
347 If the total utilization U=sum(WCET_i/P_i) is larger than M (with M equal
348 to the number of CPUs), then the scheduler is unable to respect all the
349 deadlines.
350 Note that total utilization is defined as the sum of the utilizations
351 WCET_i/P_i over all the real-time tasks in the system. When considering
352 multiple real-time tasks, the parameters of the i-th task are indicated
353 with the "_i" suffix.
354 Moreover, if the total utilization is larger than M, then we risk starving
355 non- real-time tasks by real-time tasks.
356 If, instead, the total utilization is smaller than M, then non real-time
357 tasks will not be starved and the system might be able to respect all the
358 deadlines.
359 As a matter of fact, in this case it is possible to provide an upper bound
360 for tardiness (defined as the maximum between 0 and the difference
361 between the finishing time of a job and its absolute deadline).
362 More precisely, it can be proven that using a global EDF scheduler the
363 maximum tardiness of each task is smaller or equal than
364
365 ((M − 1) · WCET_max − WCET_min)/(M − (M − 2) · U_max) + WCET_max
366
367 where WCET_max = max{WCET_i} is the maximum WCET, WCET_min=min{WCET_i}
368 is the minimum WCET, and U_max = max{WCET_i/P_i} is the maximum
369 utilization[12].
370
371 3.2 Schedulability Analysis for Uniprocessor Systems
372 ----------------------------------------------------
373
374 If M=1 (uniprocessor system), or in case of partitioned scheduling (each
375 real-time task is statically assigned to one and only one CPU), it is
376 possible to formally check if all the deadlines are respected.
377 If D_i = P_i for all tasks, then EDF is able to respect all the deadlines
378 of all the tasks executing on a CPU if and only if the total utilization
379 of the tasks running on such a CPU is smaller or equal than 1.
380 If D_i != P_i for some task, then it is possible to define the density of
381 a task as WCET_i/min{D_i,P_i}, and EDF is able to respect all the deadlines
382 of all the tasks running on a CPU if the sum of the densities of the tasks
383 running on such a CPU is smaller or equal than 1:
384
385 sum(WCET_i / min{D_i, P_i}) <= 1
386
387 It is important to notice that this condition is only sufficient, and not
388 necessary: there are task sets that are schedulable, but do not respect the
389 condition. For example, consider the task set {Task_1,Task_2} composed by
390 Task_1=(50ms,50ms,100ms) and Task_2=(10ms,100ms,100ms).
391 EDF is clearly able to schedule the two tasks without missing any deadline
392 (Task_1 is scheduled as soon as it is released, and finishes just in time
393 to respect its deadline; Task_2 is scheduled immediately after Task_1, hence
394 its response time cannot be larger than 50ms + 10ms = 60ms) even if
395
396 50 / min{50,100} + 10 / min{100, 100} = 50 / 50 + 10 / 100 = 1.1
397
398 Of course it is possible to test the exact schedulability of tasks with
399 D_i != P_i (checking a condition that is both sufficient and necessary),
400 but this cannot be done by comparing the total utilization or density with
401 a constant. Instead, the so called "processor demand" approach can be used,
402 computing the total amount of CPU time h(t) needed by all the tasks to
403 respect all of their deadlines in a time interval of size t, and comparing
404 such a time with the interval size t. If h(t) is smaller than t (that is,
405 the amount of time needed by the tasks in a time interval of size t is
406 smaller than the size of the interval) for all the possible values of t, then
407 EDF is able to schedule the tasks respecting all of their deadlines. Since
408 performing this check for all possible values of t is impossible, it has been
409 proven[4,5,6] that it is sufficient to perform the test for values of t
410 between 0 and a maximum value L. The cited papers contain all of the
411 mathematical details and explain how to compute h(t) and L.
412 In any case, this kind of analysis is too complex as well as too
413 time-consuming to be performed on-line. Hence, as explained in Section
414 4 Linux uses an admission test based on the tasks' utilizations.
415
416 3.3 Schedulability Analysis for Multiprocessor Systems
417 ------------------------------------------------------
418
419 On multiprocessor systems with global EDF scheduling (non partitioned
420 systems), a sufficient test for schedulability can not be based on the
421 utilizations or densities: it can be shown that even if D_i = P_i task
422 sets with utilizations slightly larger than 1 can miss deadlines regardless
423 of the number of CPUs.
424
425 Consider a set {Task_1,...Task_{M+1}} of M+1 tasks on a system with M
426 CPUs, with the first task Task_1=(P,P,P) having period, relative deadline
427 and WCET equal to P. The remaining M tasks Task_i=(e,P-1,P-1) have an
428 arbitrarily small worst case execution time (indicated as "e" here) and a
429 period smaller than the one of the first task. Hence, if all the tasks
430 activate at the same time t, global EDF schedules these M tasks first
431 (because their absolute deadlines are equal to t + P - 1, hence they are
432 smaller than the absolute deadline of Task_1, which is t + P). As a
433 result, Task_1 can be scheduled only at time t + e, and will finish at
434 time t + e + P, after its absolute deadline. The total utilization of the
435 task set is U = M · e / (P - 1) + P / P = M · e / (P - 1) + 1, and for small
436 values of e this can become very close to 1. This is known as "Dhall's
437 effect"[7]. Note: the example in the original paper by Dhall has been
438 slightly simplified here (for example, Dhall more correctly computed
439 lim_{e->0}U).
440
441 More complex schedulability tests for global EDF have been developed in
442 real-time literature[8,9], but they are not based on a simple comparison
443 between total utilization (or density) and a fixed constant. If all tasks
444 have D_i = P_i, a sufficient schedulability condition can be expressed in
445 a simple way:
446
447 sum(WCET_i / P_i) <= M - (M - 1) · U_max
448
449 where U_max = max{WCET_i / P_i}[10]. Notice that for U_max = 1,
450 M - (M - 1) · U_max becomes M - M + 1 = 1 and this schedulability condition
451 just confirms the Dhall's effect. A more complete survey of the literature
452 about schedulability tests for multi-processor real-time scheduling can be
453 found in [11].
454
455 As seen, enforcing that the total utilization is smaller than M does not
456 guarantee that global EDF schedules the tasks without missing any deadline
457 (in other words, global EDF is not an optimal scheduling algorithm). However,
458 a total utilization smaller than M is enough to guarantee that non real-time
459 tasks are not starved and that the tardiness of real-time tasks has an upper
460 bound[12] (as previously noted). Different bounds on the maximum tardiness
461 experienced by real-time tasks have been developed in various papers[13,14],
462 but the theoretical result that is important for SCHED_DEADLINE is that if
463 the total utilization is smaller or equal than M then the response times of
464 the tasks are limited.
465
466 3.4 Relationship with SCHED_DEADLINE Parameters
467 -----------------------------------------------
468
469 Finally, it is important to understand the relationship between the
470 SCHED_DEADLINE scheduling parameters described in Section 2 (runtime,
471 deadline and period) and the real-time task parameters (WCET, D, P)
472 described in this section. Note that the tasks' temporal constraints are
473 represented by its absolute deadlines d_j = r_j + D described above, while
474 SCHED_DEADLINE schedules the tasks according to scheduling deadlines (see
475 Section 2).
476 If an admission test is used to guarantee that the scheduling deadlines
477 are respected, then SCHED_DEADLINE can be used to schedule real-time tasks
478 guaranteeing that all the jobs' deadlines of a task are respected.
479 In order to do this, a task must be scheduled by setting:
480
481 - runtime >= WCET
482 - deadline = D
483 - period <= P
484
485 IOW, if runtime >= WCET and if period is <= P, then the scheduling deadlines
486 and the absolute deadlines (d_j) coincide, so a proper admission control
487 allows to respect the jobs' absolute deadlines for this task (this is what is
488 called "hard schedulability property" and is an extension of Lemma 1 of [2]).
489 Notice that if runtime > deadline the admission control will surely reject
490 this task, as it is not possible to respect its temporal constraints.
491
492 References:
493
494 1 - C. L. Liu and J. W. Layland. Scheduling algorithms for multiprogram-
495 ming in a hard-real-time environment. Journal of the Association for
496 Computing Machinery, 20(1), 1973.
497 2 - L. Abeni , G. Buttazzo. Integrating Multimedia Applications in Hard
498 Real-Time Systems. Proceedings of the 19th IEEE Real-time Systems
499 Symposium, 1998. http://retis.sssup.it/~giorgio/paps/1998/rtss98-cbs.pdf
500 3 - L. Abeni. Server Mechanisms for Multimedia Applications. ReTiS Lab
501 Technical Report. http://disi.unitn.it/~abeni/tr-98-01.pdf
502 4 - J. Y. Leung and M.L. Merril. A Note on Preemptive Scheduling of
503 Periodic, Real-Time Tasks. Information Processing Letters, vol. 11,
504 no. 3, pp. 115-118, 1980.
505 5 - S. K. Baruah, A. K. Mok and L. E. Rosier. Preemptively Scheduling
506 Hard-Real-Time Sporadic Tasks on One Processor. Proceedings of the
507 11th IEEE Real-time Systems Symposium, 1990.
508 6 - S. K. Baruah, L. E. Rosier and R. R. Howell. Algorithms and Complexity
509 Concerning the Preemptive Scheduling of Periodic Real-Time tasks on
510 One Processor. Real-Time Systems Journal, vol. 4, no. 2, pp 301-324,
511 1990.
512 7 - S. J. Dhall and C. L. Liu. On a real-time scheduling problem. Operations
513 research, vol. 26, no. 1, pp 127-140, 1978.
514 8 - T. Baker. Multiprocessor EDF and Deadline Monotonic Schedulability
515 Analysis. Proceedings of the 24th IEEE Real-Time Systems Symposium, 2003.
516 9 - T. Baker. An Analysis of EDF Schedulability on a Multiprocessor.
517 IEEE Transactions on Parallel and Distributed Systems, vol. 16, no. 8,
518 pp 760-768, 2005.
519 10 - J. Goossens, S. Funk and S. Baruah, Priority-Driven Scheduling of
520 Periodic Task Systems on Multiprocessors. Real-Time Systems Journal,
521 vol. 25, no. 2–3, pp. 187–205, 2003.
522 11 - R. Davis and A. Burns. A Survey of Hard Real-Time Scheduling for
523 Multiprocessor Systems. ACM Computing Surveys, vol. 43, no. 4, 2011.
524 http://www-users.cs.york.ac.uk/~robdavis/papers/MPSurveyv5.0.pdf
525 12 - U. C. Devi and J. H. Anderson. Tardiness Bounds under Global EDF
526 Scheduling on a Multiprocessor. Real-Time Systems Journal, vol. 32,
527 no. 2, pp 133-189, 2008.
528 13 - P. Valente and G. Lipari. An Upper Bound to the Lateness of Soft
529 Real-Time Tasks Scheduled by EDF on Multiprocessors. Proceedings of
530 the 26th IEEE Real-Time Systems Symposium, 2005.
531 14 - J. Erickson, U. Devi and S. Baruah. Improved tardiness bounds for
532 Global EDF. Proceedings of the 22nd Euromicro Conference on
533 Real-Time Systems, 2010.
534 15 - G. Lipari, S. Baruah, Greedy reclamation of unused bandwidth in
535 constant-bandwidth servers, 12th IEEE Euromicro Conference on Real-Time
536 Systems, 2000.
537 16 - L. Abeni, J. Lelli, C. Scordino, L. Palopoli, Greedy CPU reclaiming for
538 SCHED DEADLINE. In Proceedings of the Real-Time Linux Workshop (RTLWS),
539 Dusseldorf, Germany, 2014.
540 17 - L. Abeni, G. Lipari, A. Parri, Y. Sun, Multicore CPU reclaiming: parallel
541 or sequential?. In Proceedings of the 31st Annual ACM Symposium on Applied
542 Computing, 2016.
543 18 - J. Lelli, C. Scordino, L. Abeni, D. Faggioli, Deadline scheduling in the
544 Linux kernel, Software: Practice and Experience, 46(6): 821-839, June
545 2016.
546 19 - C. Scordino, L. Abeni, J. Lelli, Energy-Aware Real-Time Scheduling in
547 the Linux Kernel, 33rd ACM/SIGAPP Symposium On Applied Computing (SAC
548 2018), Pau, France, April 2018.
549
550
551 4. Bandwidth management
552 =======================
553
554 As previously mentioned, in order for -deadline scheduling to be
555 effective and useful (that is, to be able to provide "runtime" time units
556 within "deadline"), it is important to have some method to keep the allocation
557 of the available fractions of CPU time to the various tasks under control.
558 This is usually called "admission control" and if it is not performed, then
559 no guarantee can be given on the actual scheduling of the -deadline tasks.
560
561 As already stated in Section 3, a necessary condition to be respected to
562 correctly schedule a set of real-time tasks is that the total utilization
563 is smaller than M. When talking about -deadline tasks, this requires that
564 the sum of the ratio between runtime and period for all tasks is smaller
565 than M. Notice that the ratio runtime/period is equivalent to the utilization
566 of a "traditional" real-time task, and is also often referred to as
567 "bandwidth".
568 The interface used to control the CPU bandwidth that can be allocated
569 to -deadline tasks is similar to the one already used for -rt
570 tasks with real-time group scheduling (a.k.a. RT-throttling - see
571 Documentation/scheduler/sched-rt-group.rst), and is based on readable/
572 writable control files located in procfs (for system wide settings).
573 Notice that per-group settings (controlled through cgroupfs) are still not
574 defined for -deadline tasks, because more discussion is needed in order to
575 figure out how we want to manage SCHED_DEADLINE bandwidth at the task group
576 level.
577
578 A main difference between deadline bandwidth management and RT-throttling
579 is that -deadline tasks have bandwidth on their own (while -rt ones don't!),
580 and thus we don't need a higher level throttling mechanism to enforce the
581 desired bandwidth. In other words, this means that interface parameters are
582 only used at admission control time (i.e., when the user calls
583 sched_setattr()). Scheduling is then performed considering actual tasks'
584 parameters, so that CPU bandwidth is allocated to SCHED_DEADLINE tasks
585 respecting their needs in terms of granularity. Therefore, using this simple
586 interface we can put a cap on total utilization of -deadline tasks (i.e.,
587 \Sum (runtime_i / period_i) < global_dl_utilization_cap).
588
589 4.1 System wide settings
590 ------------------------
591
592 The system wide settings are configured under the /proc virtual file system.
593
594 For now the -rt knobs are used for -deadline admission control and the
595 -deadline runtime is accounted against the -rt runtime. We realize that this
596 isn't entirely desirable; however, it is better to have a small interface for
597 now, and be able to change it easily later. The ideal situation (see 5.) is to
598 run -rt tasks from a -deadline server; in which case the -rt bandwidth is a
599 direct subset of dl_bw.
600
601 This means that, for a root_domain comprising M CPUs, -deadline tasks
602 can be created while the sum of their bandwidths stays below:
603
604 M * (sched_rt_runtime_us / sched_rt_period_us)
605
606 It is also possible to disable this bandwidth management logic, and
607 be thus free of oversubscribing the system up to any arbitrary level.
608 This is done by writing -1 in /proc/sys/kernel/sched_rt_runtime_us.
609
610
611 4.2 Task interface
612 ------------------
613
614 Specifying a periodic/sporadic task that executes for a given amount of
615 runtime at each instance, and that is scheduled according to the urgency of
616 its own timing constraints needs, in general, a way of declaring:
617
618 - a (maximum/typical) instance execution time,
619 - a minimum interval between consecutive instances,
620 - a time constraint by which each instance must be completed.
621
622 Therefore:
623
624 * a new struct sched_attr, containing all the necessary fields is
625 provided;
626 * the new scheduling related syscalls that manipulate it, i.e.,
627 sched_setattr() and sched_getattr() are implemented.
628
629 For debugging purposes, the leftover runtime and absolute deadline of a
630 SCHED_DEADLINE task can be retrieved through /proc/<pid>/sched (entries
631 dl.runtime and dl.deadline, both values in ns). A programmatic way to
632 retrieve these values from production code is under discussion.
633
634
635 4.3 Default behavior
636 ---------------------
637
638 The default value for SCHED_DEADLINE bandwidth is to have rt_runtime equal to
639 950000. With rt_period equal to 1000000, by default, it means that -deadline
640 tasks can use at most 95%, multiplied by the number of CPUs that compose the
641 root_domain, for each root_domain.
642 This means that non -deadline tasks will receive at least 5% of the CPU time,
643 and that -deadline tasks will receive their runtime with a guaranteed
644 worst-case delay respect to the "deadline" parameter. If "deadline" = "period"
645 and the cpuset mechanism is used to implement partitioned scheduling (see
646 Section 5), then this simple setting of the bandwidth management is able to
647 deterministically guarantee that -deadline tasks will receive their runtime
648 in a period.
649
650 Finally, notice that in order not to jeopardize the admission control a
651 -deadline task cannot fork.
652
653
654 4.4 Behavior of sched_yield()
655 -----------------------------
656
657 When a SCHED_DEADLINE task calls sched_yield(), it gives up its
658 remaining runtime and is immediately throttled, until the next
659 period, when its runtime will be replenished (a special flag
660 dl_yielded is set and used to handle correctly throttling and runtime
661 replenishment after a call to sched_yield()).
662
663 This behavior of sched_yield() allows the task to wake-up exactly at
664 the beginning of the next period. Also, this may be useful in the
665 future with bandwidth reclaiming mechanisms, where sched_yield() will
666 make the leftoever runtime available for reclamation by other
667 SCHED_DEADLINE tasks.
668
669
670 5. Tasks CPU affinity
671 =====================
672
673 -deadline tasks cannot have an affinity mask smaller that the entire
674 root_domain they are created on. However, affinities can be specified
675 through the cpuset facility (Documentation/admin-guide/cgroup-v1/cpusets.rst).
676
677 5.1 SCHED_DEADLINE and cpusets HOWTO
678 ------------------------------------
679
680 An example of a simple configuration (pin a -deadline task to CPU0)
681 follows (rt-app is used to create a -deadline task)::
682
683 mkdir /dev/cpuset
684 mount -t cgroup -o cpuset cpuset /dev/cpuset
685 cd /dev/cpuset
686 mkdir cpu0
687 echo 0 > cpu0/cpuset.cpus
688 echo 0 > cpu0/cpuset.mems
689 echo 1 > cpuset.cpu_exclusive
690 echo 0 > cpuset.sched_load_balance
691 echo 1 > cpu0/cpuset.cpu_exclusive
692 echo 1 > cpu0/cpuset.mem_exclusive
693 echo $$ > cpu0/tasks
694 rt-app -t 100000:10000:d:0 -D5 # it is now actually superfluous to specify
695 # task affinity
696
697 6. Future plans
698 ===============
699
700 Still missing:
701
702 - programmatic way to retrieve current runtime and absolute deadline
703 - refinements to deadline inheritance, especially regarding the possibility
704 of retaining bandwidth isolation among non-interacting tasks. This is
705 being studied from both theoretical and practical points of view, and
706 hopefully we should be able to produce some demonstrative code soon;
707 - (c)group based bandwidth management, and maybe scheduling;
708 - access control for non-root users (and related security concerns to
709 address), which is the best way to allow unprivileged use of the mechanisms
710 and how to prevent non-root users "cheat" the system?
711
712 As already discussed, we are planning also to merge this work with the EDF
713 throttling patches [https://lore.kernel.org/r/cover.1266931410.git.fabio@helm.retis] but we still are in
714 the preliminary phases of the merge and we really seek feedback that would
715 help us decide on the direction it should take.
716
717 Appendix A. Test suite
718 ======================
719
720 The SCHED_DEADLINE policy can be easily tested using two applications that
721 are part of a wider Linux Scheduler validation suite. The suite is
722 available as a GitHub repository: https://github.com/scheduler-tools.
723
724 The first testing application is called rt-app and can be used to
725 start multiple threads with specific parameters. rt-app supports
726 SCHED_{OTHER,FIFO,RR,DEADLINE} scheduling policies and their related
727 parameters (e.g., niceness, priority, runtime/deadline/period). rt-app
728 is a valuable tool, as it can be used to synthetically recreate certain
729 workloads (maybe mimicking real use-cases) and evaluate how the scheduler
730 behaves under such workloads. In this way, results are easily reproducible.
731 rt-app is available at: https://github.com/scheduler-tools/rt-app.
732
733 Thread parameters can be specified from the command line, with something like
734 this::
735
736 # rt-app -t 100000:10000:d -t 150000:20000:f:10 -D5
737
738 The above creates 2 threads. The first one, scheduled by SCHED_DEADLINE,
739 executes for 10ms every 100ms. The second one, scheduled at SCHED_FIFO
740 priority 10, executes for 20ms every 150ms. The test will run for a total
741 of 5 seconds.
742
743 More interestingly, configurations can be described with a json file that
744 can be passed as input to rt-app with something like this::
745
746 # rt-app my_config.json
747
748 The parameters that can be specified with the second method are a superset
749 of the command line options. Please refer to rt-app documentation for more
750 details (`<rt-app-sources>/doc/*.json`).
751
752 The second testing application is a modification of schedtool, called
753 schedtool-dl, which can be used to setup SCHED_DEADLINE parameters for a
754 certain pid/application. schedtool-dl is available at:
755 https://github.com/scheduler-tools/schedtool-dl.git.
756
757 The usage is straightforward::
758
759 # schedtool -E -t 10000000:100000000 -e ./my_cpuhog_app
760
761 With this, my_cpuhog_app is put to run inside a SCHED_DEADLINE reservation
762 of 10ms every 100ms (note that parameters are expressed in microseconds).
763 You can also use schedtool to create a reservation for an already running
764 application, given that you know its pid::
765
766 # schedtool -E -t 10000000:100000000 my_app_pid
767
768 Appendix B. Minimal main()
769 ==========================
770
771 We provide in what follows a simple (ugly) self-contained code snippet
772 showing how SCHED_DEADLINE reservations can be created by a real-time
773 application developer::
774
775 #define _GNU_SOURCE
776 #include <unistd.h>
777 #include <stdio.h>
778 #include <stdlib.h>
779 #include <string.h>
780 #include <time.h>
781 #include <linux/unistd.h>
782 #include <linux/kernel.h>
783 #include <linux/types.h>
784 #include <sys/syscall.h>
785 #include <pthread.h>
786
787 #define gettid() syscall(__NR_gettid)
788
789 #define SCHED_DEADLINE 6
790
791 /* XXX use the proper syscall numbers */
792 #ifdef __x86_64__
793 #define __NR_sched_setattr 314
794 #define __NR_sched_getattr 315
795 #endif
796
797 #ifdef __i386__
798 #define __NR_sched_setattr 351
799 #define __NR_sched_getattr 352
800 #endif
801
802 #ifdef __arm__
803 #define __NR_sched_setattr 380
804 #define __NR_sched_getattr 381
805 #endif
806
807 static volatile int done;
808
809 struct sched_attr {
810 __u32 size;
811
812 __u32 sched_policy;
813 __u64 sched_flags;
814
815 /* SCHED_NORMAL, SCHED_BATCH */
816 __s32 sched_nice;
817
818 /* SCHED_FIFO, SCHED_RR */
819 __u32 sched_priority;
820
821 /* SCHED_DEADLINE (nsec) */
822 __u64 sched_runtime;
823 __u64 sched_deadline;
824 __u64 sched_period;
825 };
826
827 int sched_setattr(pid_t pid,
828 const struct sched_attr *attr,
829 unsigned int flags)
830 {
831 return syscall(__NR_sched_setattr, pid, attr, flags);
832 }
833
834 int sched_getattr(pid_t pid,
835 struct sched_attr *attr,
836 unsigned int size,
837 unsigned int flags)
838 {
839 return syscall(__NR_sched_getattr, pid, attr, size, flags);
840 }
841
842 void *run_deadline(void *data)
843 {
844 struct sched_attr attr;
845 int x = 0;
846 int ret;
847 unsigned int flags = 0;
848
849 printf("deadline thread started [%ld]\n", gettid());
850
851 attr.size = sizeof(attr);
852 attr.sched_flags = 0;
853 attr.sched_nice = 0;
854 attr.sched_priority = 0;
855
856 /* This creates a 10ms/30ms reservation */
857 attr.sched_policy = SCHED_DEADLINE;
858 attr.sched_runtime = 10 * 1000 * 1000;
859 attr.sched_period = attr.sched_deadline = 30 * 1000 * 1000;
860
861 ret = sched_setattr(0, &attr, flags);
862 if (ret < 0) {
863 done = 0;
864 perror("sched_setattr");
865 exit(-1);
866 }
867
868 while (!done) {
869 x++;
870 }
871
872 printf("deadline thread dies [%ld]\n", gettid());
873 return NULL;
874 }
875
876 int main (int argc, char **argv)
877 {
878 pthread_t thread;
879
880 printf("main thread [%ld]\n", gettid());
881
882 pthread_create(&thread, NULL, run_deadline, NULL);
883
884 sleep(10);
885
886 done = 1;
887 pthread_join(thread, NULL);
888
889 printf("main dies [%ld]\n", gettid());
890 return 0;
891 }
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.