2 * Copyright (c) 2009-2016 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
29 #include <mach/mach_types.h>
30 #include <mach/machine.h>
31 #include <mach/policy.h>
32 #include <mach/sync_policy.h>
33 #include <mach/thread_act.h>
35 #include <machine/machine_routines.h>
36 #include <machine/sched_param.h>
37 #include <machine/machine_cpu.h>
39 #include <kern/kern_types.h>
40 #include <kern/clock.h>
41 #include <kern/counters.h>
42 #include <kern/cpu_number.h>
43 #include <kern/cpu_data.h>
44 #include <kern/debug.h>
45 #include <kern/macro_help.h>
46 #include <kern/machine.h>
47 #include <kern/misc_protos.h>
48 #include <kern/processor.h>
49 #include <kern/queue.h>
50 #include <kern/sched.h>
51 #include <kern/sched_prim.h>
52 #include <kern/syscall_subr.h>
53 #include <kern/task.h>
54 #include <kern/thread.h>
57 #include <vm/vm_kern.h>
58 #include <vm/vm_map.h>
62 #include <sys/kdebug.h>
64 #if defined(CONFIG_SCHED_GRRR_CORE)
67 grrr_priority_mapping_init(void);
85 grrr_sorted_list_insert_group(grrr_run_queue_t rq
,
89 grrr_rescale_work(grrr_run_queue_t rq
);
92 grrr_runqueue_init(grrr_run_queue_t runq
);
94 /* Map Mach priorities to ones suitable for proportional sharing */
95 static grrr_proportional_priority_t grrr_priority_mapping
[NRQS
];
97 /* Map each proportional priority to its group */
98 static grrr_group_index_t grrr_group_mapping
[NUM_GRRR_PROPORTIONAL_PRIORITIES
];
100 uint32_t grrr_rescale_tick
;
102 #endif /* defined(CONFIG_SCHED_GRRR_CORE) */
104 #if defined(CONFIG_SCHED_GRRR)
107 sched_grrr_init(void);
110 sched_grrr_timebase_init(void);
113 sched_grrr_processor_init(processor_t processor
);
116 sched_grrr_pset_init(processor_set_t pset
);
119 sched_grrr_maintenance_continuation(void);
122 sched_grrr_choose_thread(processor_t processor
,
127 sched_grrr_steal_thread(processor_set_t pset
);
130 sched_grrr_compute_priority(thread_t thread
);
133 sched_grrr_choose_processor( processor_set_t pset
,
134 processor_t processor
,
138 sched_grrr_processor_enqueue(
139 processor_t processor
,
144 sched_grrr_processor_queue_shutdown(
145 processor_t processor
);
148 sched_grrr_processor_queue_remove(
149 processor_t processor
,
153 sched_grrr_processor_queue_empty(processor_t processor
);
156 sched_grrr_processor_queue_has_priority(processor_t processor
,
161 sched_grrr_priority_is_urgent(int priority
);
164 sched_grrr_processor_csw_check(processor_t processor
);
167 sched_grrr_initial_quantum_size(thread_t thread
);
170 sched_grrr_initial_thread_sched_mode(task_t parent_task
);
173 sched_grrr_can_update_priority(thread_t thread
);
176 sched_grrr_update_priority(thread_t thread
);
179 sched_grrr_lightweight_update_priority(thread_t thread
);
182 sched_grrr_processor_runq_count(processor_t processor
);
185 sched_grrr_processor_runq_stats_count_sum(processor_t processor
);
188 sched_grrr_processor_bound_count(processor_t processor
);
191 sched_grrr_thread_update_scan(sched_update_scan_context_t scan_context
);
193 const struct sched_dispatch_table sched_grrr_dispatch
= {
194 .sched_name
= "grrr",
195 .init
= sched_grrr_init
,
196 .timebase_init
= sched_grrr_timebase_init
,
197 .processor_init
= sched_grrr_processor_init
,
198 .pset_init
= sched_grrr_pset_init
,
199 .maintenance_continuation
= sched_grrr_maintenance_continuation
,
200 .choose_thread
= sched_grrr_choose_thread
,
201 .steal_thread_enabled
= FALSE
,
202 .steal_thread
= sched_grrr_steal_thread
,
203 .compute_timeshare_priority
= sched_grrr_compute_priority
,
204 .choose_processor
= sched_grrr_choose_processor
,
205 .processor_enqueue
= sched_grrr_processor_enqueue
,
206 .processor_queue_shutdown
= sched_grrr_processor_queue_shutdown
,
207 .processor_queue_remove
= sched_grrr_processor_queue_remove
,
208 .processor_queue_empty
= sched_grrr_processor_queue_empty
,
209 .priority_is_urgent
= sched_grrr_priority_is_urgent
,
210 .processor_csw_check
= sched_grrr_processor_csw_check
,
211 .processor_queue_has_priority
= sched_grrr_processor_queue_has_priority
,
212 .initial_quantum_size
= sched_grrr_initial_quantum_size
,
213 .initial_thread_sched_mode
= sched_grrr_initial_thread_sched_mode
,
214 .can_update_priority
= sched_grrr_can_update_priority
,
215 .update_priority
= sched_grrr_update_priority
,
216 .lightweight_update_priority
= sched_grrr_lightweight_update_priority
,
217 .quantum_expire
= sched_default_quantum_expire
,
218 .processor_runq_count
= sched_grrr_processor_runq_count
,
219 .processor_runq_stats_count_sum
= sched_grrr_processor_runq_stats_count_sum
,
220 .processor_bound_count
= sched_grrr_processor_bound_count
,
221 .thread_update_scan
= sched_grrr_thread_update_scan
,
222 .direct_dispatch_to_idle_processors
= TRUE
,
223 .multiple_psets_enabled
= TRUE
,
224 .sched_groups_enabled
= FALSE
,
225 .avoid_processor_enabled
= FALSE
,
226 .thread_avoid_processor
= NULL
,
227 .processor_balance
= sched_SMT_balance
,
229 .rt_runq
= sched_rtglobal_runq
,
230 .rt_init
= sched_rtglobal_init
,
231 .rt_queue_shutdown
= sched_rtglobal_queue_shutdown
,
232 .rt_runq_scan
= sched_rtglobal_runq_scan
,
233 .rt_runq_count_sum
= sched_rtglobal_runq_count_sum
,
235 .qos_max_parallelism
= sched_qos_max_parallelism
,
236 .check_spill
= sched_check_spill
,
237 .ipi_policy
= sched_ipi_policy
,
238 .thread_should_yield
= sched_thread_should_yield
,
241 extern int max_unsafe_quanta
;
243 static uint32_t grrr_quantum_us
;
244 static uint32_t grrr_quantum
;
246 static uint64_t sched_grrr_tick_deadline
;
249 sched_grrr_init(void)
251 if (default_preemption_rate
< 1)
252 default_preemption_rate
= 100;
253 grrr_quantum_us
= (1000 * 1000) / default_preemption_rate
;
255 printf("standard grrr timeslicing quantum is %d us\n", grrr_quantum_us
);
257 grrr_priority_mapping_init();
261 sched_grrr_timebase_init(void)
265 /* standard timeslicing quantum */
266 clock_interval_to_absolutetime_interval(
267 grrr_quantum_us
, NSEC_PER_USEC
, &abstime
);
268 assert((abstime
>> 32) == 0 && (uint32_t)abstime
!= 0);
269 grrr_quantum
= (uint32_t)abstime
;
271 thread_depress_time
= 1 * grrr_quantum
;
272 default_timeshare_computation
= grrr_quantum
/ 2;
273 default_timeshare_constraint
= grrr_quantum
;
275 max_unsafe_computation
= max_unsafe_quanta
* grrr_quantum
;
276 sched_safe_duration
= 2 * max_unsafe_quanta
* grrr_quantum
;
281 sched_grrr_processor_init(processor_t processor
)
283 grrr_runqueue_init(&processor
->grrr_runq
);
287 sched_grrr_pset_init(processor_set_t pset __unused
)
292 sched_grrr_maintenance_continuation(void)
294 uint64_t abstime
= mach_absolute_time();
299 * Compute various averages.
303 if (sched_grrr_tick_deadline
== 0)
304 sched_grrr_tick_deadline
= abstime
;
306 clock_deadline_for_periodic_event(10*sched_one_second_interval
, abstime
,
307 &sched_grrr_tick_deadline
);
309 assert_wait_deadline((event_t
)sched_grrr_maintenance_continuation
, THREAD_UNINT
, sched_grrr_tick_deadline
);
310 thread_block((thread_continue_t
)sched_grrr_maintenance_continuation
);
315 sched_grrr_choose_thread(processor_t processor
,
316 int priority __unused
,
317 ast_t reason __unused
)
319 grrr_run_queue_t rq
= &processor
->grrr_runq
;
321 return grrr_select(rq
);
325 sched_grrr_steal_thread(processor_set_t pset
)
333 sched_grrr_compute_priority(thread_t thread
)
335 return thread
->base_pri
;
339 sched_grrr_choose_processor( processor_set_t pset
,
340 processor_t processor
,
343 return choose_processor(pset
, processor
, thread
);
347 sched_grrr_processor_enqueue(
348 processor_t processor
,
350 integer_t options __unused
)
352 grrr_run_queue_t rq
= &processor
->grrr_runq
;
355 result
= grrr_enqueue(rq
, thread
);
357 thread
->runq
= processor
;
363 sched_grrr_processor_queue_shutdown(
364 processor_t processor
)
366 processor_set_t pset
= processor
->processor_set
;
368 queue_head_t tqueue
, bqueue
;
373 while ((thread
= sched_grrr_choose_thread(processor
, IDLEPRI
, AST_NONE
)) != THREAD_NULL
) {
374 if (thread
->bound_processor
== PROCESSOR_NULL
) {
375 enqueue_tail(&tqueue
, (queue_entry_t
)thread
);
377 enqueue_tail(&bqueue
, (queue_entry_t
)thread
);
381 while ((thread
= (thread_t
)(void *)dequeue_head(&bqueue
)) != THREAD_NULL
) {
382 sched_grrr_processor_enqueue(processor
, thread
, SCHED_TAILQ
);
387 while ((thread
= (thread_t
)(void *)dequeue_head(&tqueue
)) != THREAD_NULL
) {
390 thread_setrun(thread
, SCHED_TAILQ
);
392 thread_unlock(thread
);
397 sched_grrr_processor_queue_remove(
398 processor_t processor
,
401 processor_set_t pset
= processor
->processor_set
;
405 if (processor
== thread
->runq
) {
407 * Thread is on a run queue and we have a lock on
410 grrr_run_queue_t rq
= &processor
->grrr_runq
;
412 grrr_remove(rq
, thread
);
415 * The thread left the run queue before we could
416 * lock the run queue.
418 assert(thread
->runq
== PROCESSOR_NULL
);
419 processor
= PROCESSOR_NULL
;
424 return (processor
!= PROCESSOR_NULL
);
428 sched_grrr_processor_queue_empty(processor_t processor __unused
)
432 result
= (processor
->grrr_runq
.count
== 0);
438 sched_grrr_processor_queue_has_priority(processor_t processor
,
440 boolean_t gte __unused
)
442 grrr_run_queue_t rq
= &processor
->grrr_runq
;
445 i
= grrr_group_mapping
[grrr_priority_mapping
[priority
]];
446 for ( ; i
< NUM_GRRR_GROUPS
; i
++) {
447 if (rq
->groups
[i
].count
> 0)
454 /* Implement sched_preempt_pri in code */
456 sched_grrr_priority_is_urgent(int priority
)
458 if (priority
<= BASEPRI_FOREGROUND
)
461 if (priority
< MINPRI_KERNEL
)
464 if (priority
>= BASEPRI_PREEMPT
)
471 sched_grrr_processor_csw_check(processor_t processor
)
475 count
= sched_grrr_processor_runq_count(processor
);
484 sched_grrr_initial_quantum_size(thread_t thread __unused
)
490 sched_grrr_initial_thread_sched_mode(task_t parent_task
)
492 if (parent_task
== kernel_task
)
493 return TH_MODE_FIXED
;
495 return TH_MODE_TIMESHARE
;
499 sched_grrr_can_update_priority(thread_t thread __unused
)
505 sched_grrr_update_priority(thread_t thread __unused
)
511 sched_grrr_lightweight_update_priority(thread_t thread __unused
)
517 sched_grrr_processor_runq_count(processor_t processor
)
519 return processor
->grrr_runq
.count
;
523 sched_grrr_processor_runq_stats_count_sum(processor_t processor
)
525 return processor
->grrr_runq
.runq_stats
.count_sum
;
529 sched_grrr_processor_bound_count(__unused processor_t processor
)
535 sched_grrr_thread_update_scan(__unused sched_update_scan_context_t scan_context
)
540 #endif /* defined(CONFIG_SCHED_GRRR) */
542 #if defined(CONFIG_SCHED_GRRR_CORE)
545 grrr_priority_mapping_init(void)
549 /* Map 0->0 up to 10->20 */
550 for (i
=0; i
<= 10; i
++) {
551 grrr_priority_mapping
[i
] = 2*i
;
554 /* Map user priorities 11->33 up to 51 -> 153 */
555 for (i
=11; i
<= 51; i
++) {
556 grrr_priority_mapping
[i
] = 3*i
;
559 /* Map high priorities 52->180 up to 127->255 */
560 for (i
=52; i
<= 127; i
++) {
561 grrr_priority_mapping
[i
] = 128 + i
;
564 for (i
= 0; i
< NUM_GRRR_PROPORTIONAL_PRIORITIES
; i
++) {
568 /* Calculate log(i); */
569 for (j
=0, k
=1; k
<= i
; j
++, k
*= 2);
573 grrr_group_mapping
[i
] = i
>> 2;
578 grrr_intragroup_schedule(grrr_group_t group
)
582 if (group
->count
== 0) {
586 thread
= group
->current_client
;
587 if (thread
== THREAD_NULL
) {
588 thread
= (thread_t
)(void *)queue_first(&group
->clients
);
591 if (1 /* deficit */) {
592 group
->current_client
= (thread_t
)(void *)queue_next((queue_entry_t
)thread
);
593 if (queue_end(&group
->clients
, (queue_entry_t
)group
->current_client
)) {
594 group
->current_client
= (thread_t
)(void *)queue_first(&group
->clients
);
597 thread
= group
->current_client
;
604 grrr_intergroup_schedule(grrr_run_queue_t rq
)
609 if (rq
->count
== 0) {
613 group
= rq
->current_group
;
615 if (group
== GRRR_GROUP_NULL
) {
616 group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
619 thread
= grrr_intragroup_schedule(group
);
621 if ((group
->work
>= (UINT32_MAX
-256)) || (rq
->last_rescale_tick
!= grrr_rescale_tick
)) {
622 grrr_rescale_work(rq
);
626 if (queue_end(&rq
->sorted_group_list
, queue_next((queue_entry_t
)group
))) {
627 /* last group, go back to beginning */
628 group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
630 grrr_group_t nextgroup
= (grrr_group_t
)queue_next((queue_entry_t
)group
);
631 uint64_t orderleft
, orderright
;
634 * The well-ordering condition for intergroup selection is:
636 * (group->work+1) / (nextgroup->work+1) > (group->weight) / (nextgroup->weight)
638 * Multiply both sides by their denominators to avoid division
641 orderleft
= (group
->work
+ 1) * ((uint64_t)nextgroup
->weight
);
642 orderright
= (nextgroup
->work
+ 1) * ((uint64_t)group
->weight
);
643 if (orderleft
> orderright
) {
646 group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
650 rq
->current_group
= group
;
656 grrr_runqueue_init(grrr_run_queue_t runq
)
658 grrr_group_index_t index
;
662 for (index
= 0; index
< NUM_GRRR_GROUPS
; index
++) {
663 unsigned int prisearch
;
666 prisearch
< NUM_GRRR_PROPORTIONAL_PRIORITIES
;
668 if (grrr_group_mapping
[prisearch
] == index
) {
669 runq
->groups
[index
].minpriority
= (grrr_proportional_priority_t
)prisearch
;
674 runq
->groups
[index
].index
= index
;
676 queue_init(&runq
->groups
[index
].clients
);
677 runq
->groups
[index
].count
= 0;
678 runq
->groups
[index
].weight
= 0;
679 runq
->groups
[index
].work
= 0;
680 runq
->groups
[index
].current_client
= THREAD_NULL
;
683 queue_init(&runq
->sorted_group_list
);
685 runq
->current_group
= GRRR_GROUP_NULL
;
689 grrr_rescale_work(grrr_run_queue_t rq
)
691 grrr_group_index_t index
;
693 /* avoid overflow by scaling by 1/8th */
694 for (index
= 0; index
< NUM_GRRR_GROUPS
; index
++) {
695 rq
->groups
[index
].work
>>= 3;
698 rq
->last_rescale_tick
= grrr_rescale_tick
;
706 grrr_proportional_priority_t gpriority
;
707 grrr_group_index_t gindex
;
710 gpriority
= grrr_priority_mapping
[thread
->sched_pri
];
711 gindex
= grrr_group_mapping
[gpriority
];
712 group
= &rq
->groups
[gindex
];
715 thread
->grrr_deficit
= 0;
718 if (group
->count
== 0) {
719 /* Empty group, this is the first client */
720 enqueue_tail(&group
->clients
, (queue_entry_t
)thread
);
722 group
->weight
= gpriority
;
723 group
->current_client
= thread
;
725 /* Insert before the current client */
726 if (group
->current_client
== THREAD_NULL
||
727 queue_first(&group
->clients
) == (queue_entry_t
)group
->current_client
) {
728 enqueue_head(&group
->clients
, (queue_entry_t
)thread
);
730 insque((queue_entry_t
)thread
, queue_prev((queue_entry_t
)group
->current_client
));
732 SCHED_STATS_RUNQ_CHANGE(&rq
->runq_stats
, rq
->count
);
734 group
->weight
+= gpriority
;
736 /* Since there was already a client, this is on the per-processor sorted list already */
737 remqueue((queue_entry_t
)group
);
740 grrr_sorted_list_insert_group(rq
, group
);
743 rq
->weight
+= gpriority
;
749 grrr_select(grrr_run_queue_t rq
)
753 thread
= grrr_intergroup_schedule(rq
);
754 if (thread
!= THREAD_NULL
) {
755 grrr_proportional_priority_t gpriority
;
756 grrr_group_index_t gindex
;
759 gpriority
= grrr_priority_mapping
[thread
->sched_pri
];
760 gindex
= grrr_group_mapping
[gpriority
];
761 group
= &rq
->groups
[gindex
];
763 remqueue((queue_entry_t
)thread
);
764 SCHED_STATS_RUNQ_CHANGE(&rq
->runq_stats
, rq
->count
);
766 group
->weight
-= gpriority
;
767 if (group
->current_client
== thread
) {
768 group
->current_client
= THREAD_NULL
;
771 remqueue((queue_entry_t
)group
);
772 if (group
->count
== 0) {
773 if (rq
->current_group
== group
) {
774 rq
->current_group
= GRRR_GROUP_NULL
;
777 /* Need to re-insert in sorted location */
778 grrr_sorted_list_insert_group(rq
, group
);
782 rq
->weight
-= gpriority
;
784 thread
->runq
= PROCESSOR_NULL
;
795 grrr_proportional_priority_t gpriority
;
796 grrr_group_index_t gindex
;
799 gpriority
= grrr_priority_mapping
[thread
->sched_pri
];
800 gindex
= grrr_group_mapping
[gpriority
];
801 group
= &rq
->groups
[gindex
];
803 remqueue((queue_entry_t
)thread
);
804 SCHED_STATS_RUNQ_CHANGE(&rq
->runq_stats
, rq
->count
);
806 group
->weight
-= gpriority
;
807 if (group
->current_client
== thread
) {
808 group
->current_client
= THREAD_NULL
;
811 remqueue((queue_entry_t
)group
);
812 if (group
->count
== 0) {
813 if (rq
->current_group
== group
) {
814 rq
->current_group
= GRRR_GROUP_NULL
;
817 /* Need to re-insert in sorted location */
818 grrr_sorted_list_insert_group(rq
, group
);
822 rq
->weight
-= gpriority
;
824 thread
->runq
= PROCESSOR_NULL
;
828 grrr_sorted_list_insert_group(grrr_run_queue_t rq
,
831 /* Simple insertion sort */
832 if (queue_empty(&rq
->sorted_group_list
)) {
833 enqueue_tail(&rq
->sorted_group_list
, (queue_entry_t
)group
);
835 grrr_group_t search_group
;
837 /* Start searching from the head (heaviest weight) for the first
838 * element less than us, so we can insert before it
840 search_group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
841 while (!queue_end(&rq
->sorted_group_list
, (queue_entry_t
)search_group
) ) {
843 if (search_group
->weight
< group
->weight
) {
844 /* we should be before this */
845 search_group
= (grrr_group_t
)queue_prev((queue_entry_t
)search_group
);
847 } if (search_group
->weight
== group
->weight
) {
848 /* Use group index as a tie breaker */
849 if (search_group
->index
< group
->index
) {
850 search_group
= (grrr_group_t
)queue_prev((queue_entry_t
)search_group
);
855 /* otherwise, our weight is too small, keep going */
856 search_group
= (grrr_group_t
)queue_next((queue_entry_t
)search_group
);
859 if (queue_end(&rq
->sorted_group_list
, (queue_entry_t
)search_group
)) {
860 enqueue_tail(&rq
->sorted_group_list
, (queue_entry_t
)group
);
862 insque((queue_entry_t
)group
, (queue_entry_t
)search_group
);
867 #endif /* defined(CONFIG_SCHED_GRRR_CORE) */