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
,
141 sched_options_t options
);
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
= sched_steal_thread_DISABLED
,
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 .multiple_psets_enabled
= TRUE
,
223 .sched_groups_enabled
= FALSE
,
224 .avoid_processor_enabled
= FALSE
,
225 .thread_avoid_processor
= NULL
,
226 .processor_balance
= sched_SMT_balance
,
228 .rt_runq
= sched_rtglobal_runq
,
229 .rt_init
= sched_rtglobal_init
,
230 .rt_queue_shutdown
= sched_rtglobal_queue_shutdown
,
231 .rt_runq_scan
= sched_rtglobal_runq_scan
,
232 .rt_runq_count_sum
= sched_rtglobal_runq_count_sum
,
234 .qos_max_parallelism
= sched_qos_max_parallelism
,
235 .check_spill
= sched_check_spill
,
236 .ipi_policy
= sched_ipi_policy
,
237 .thread_should_yield
= sched_thread_should_yield
,
238 .run_count_incr
= sched_run_incr
,
239 .run_count_decr
= sched_run_decr
,
240 .update_thread_bucket
= sched_update_thread_bucket
,
241 .pset_made_schedulable
= sched_pset_made_schedulable
,
244 extern int max_unsafe_quanta
;
246 static uint32_t grrr_quantum_us
;
247 static uint32_t grrr_quantum
;
249 static uint64_t sched_grrr_tick_deadline
;
252 sched_grrr_init(void)
254 if (default_preemption_rate
< 1) {
255 default_preemption_rate
= 100;
257 grrr_quantum_us
= (1000 * 1000) / default_preemption_rate
;
259 printf("standard grrr timeslicing quantum is %d us\n", grrr_quantum_us
);
261 grrr_priority_mapping_init();
265 sched_grrr_timebase_init(void)
269 /* standard timeslicing quantum */
270 clock_interval_to_absolutetime_interval(
271 grrr_quantum_us
, NSEC_PER_USEC
, &abstime
);
272 assert((abstime
>> 32) == 0 && (uint32_t)abstime
!= 0);
273 grrr_quantum
= (uint32_t)abstime
;
275 thread_depress_time
= 1 * grrr_quantum
;
276 default_timeshare_computation
= grrr_quantum
/ 2;
277 default_timeshare_constraint
= grrr_quantum
;
279 max_unsafe_computation
= max_unsafe_quanta
* grrr_quantum
;
280 sched_safe_duration
= 2 * max_unsafe_quanta
* grrr_quantum
;
284 sched_grrr_processor_init(processor_t processor
)
286 grrr_runqueue_init(&processor
->grrr_runq
);
290 sched_grrr_pset_init(processor_set_t pset __unused
)
295 sched_grrr_maintenance_continuation(void)
297 uint64_t abstime
= mach_absolute_time();
302 * Compute various averages.
306 if (sched_grrr_tick_deadline
== 0) {
307 sched_grrr_tick_deadline
= abstime
;
310 clock_deadline_for_periodic_event(10 * sched_one_second_interval
, abstime
,
311 &sched_grrr_tick_deadline
);
313 assert_wait_deadline((event_t
)sched_grrr_maintenance_continuation
, THREAD_UNINT
, sched_grrr_tick_deadline
);
314 thread_block((thread_continue_t
)sched_grrr_maintenance_continuation
);
319 sched_grrr_choose_thread(processor_t processor
,
320 int priority __unused
,
321 ast_t reason __unused
)
323 grrr_run_queue_t rq
= &processor
->grrr_runq
;
325 return grrr_select(rq
);
329 sched_grrr_steal_thread(processor_set_t pset
)
337 sched_grrr_compute_priority(thread_t thread
)
339 return thread
->base_pri
;
343 sched_grrr_choose_processor( processor_set_t pset
,
344 processor_t processor
,
347 return choose_processor(pset
, processor
, thread
);
351 sched_grrr_processor_enqueue(
352 processor_t processor
,
354 sched_options_t options __unused
)
356 grrr_run_queue_t rq
= &processor
->grrr_runq
;
359 result
= grrr_enqueue(rq
, thread
);
361 thread
->runq
= processor
;
367 sched_grrr_processor_queue_shutdown(
368 processor_t processor
)
370 processor_set_t pset
= processor
->processor_set
;
372 queue_head_t tqueue
, bqueue
;
377 while ((thread
= sched_grrr_choose_thread(processor
, IDLEPRI
, AST_NONE
)) != THREAD_NULL
) {
378 if (thread
->bound_processor
== PROCESSOR_NULL
) {
379 enqueue_tail(&tqueue
, (queue_entry_t
)thread
);
381 enqueue_tail(&bqueue
, (queue_entry_t
)thread
);
385 while ((thread
= (thread_t
)(void *)dequeue_head(&bqueue
)) != THREAD_NULL
) {
386 sched_grrr_processor_enqueue(processor
, thread
, SCHED_TAILQ
);
391 while ((thread
= (thread_t
)(void *)dequeue_head(&tqueue
)) != THREAD_NULL
) {
394 thread_setrun(thread
, SCHED_TAILQ
);
396 thread_unlock(thread
);
401 sched_grrr_processor_queue_remove(
402 processor_t processor
,
405 processor_set_t pset
= processor
->processor_set
;
409 if (processor
== thread
->runq
) {
411 * Thread is on a run queue and we have a lock on
414 grrr_run_queue_t rq
= &processor
->grrr_runq
;
416 grrr_remove(rq
, thread
);
419 * The thread left the run queue before we could
420 * lock the run queue.
422 assert(thread
->runq
== PROCESSOR_NULL
);
423 processor
= PROCESSOR_NULL
;
428 return processor
!= PROCESSOR_NULL
;
432 sched_grrr_processor_queue_empty(processor_t processor __unused
)
436 result
= (processor
->grrr_runq
.count
== 0);
442 sched_grrr_processor_queue_has_priority(processor_t processor
,
444 boolean_t gte __unused
)
446 grrr_run_queue_t rq
= &processor
->grrr_runq
;
449 i
= grrr_group_mapping
[grrr_priority_mapping
[priority
]];
450 for (; i
< NUM_GRRR_GROUPS
; i
++) {
451 if (rq
->groups
[i
].count
> 0) {
459 /* Implement sched_preempt_pri in code */
461 sched_grrr_priority_is_urgent(int priority
)
463 if (priority
<= BASEPRI_FOREGROUND
) {
467 if (priority
< MINPRI_KERNEL
) {
471 if (priority
>= BASEPRI_PREEMPT
) {
479 sched_grrr_processor_csw_check(processor_t processor
)
483 count
= sched_grrr_processor_runq_count(processor
);
493 sched_grrr_initial_quantum_size(thread_t thread __unused
)
499 sched_grrr_initial_thread_sched_mode(task_t parent_task
)
501 if (parent_task
== kernel_task
) {
502 return TH_MODE_FIXED
;
504 return TH_MODE_TIMESHARE
;
509 sched_grrr_can_update_priority(thread_t thread __unused
)
515 sched_grrr_update_priority(thread_t thread __unused
)
521 sched_grrr_lightweight_update_priority(thread_t thread __unused
)
527 sched_grrr_processor_runq_count(processor_t processor
)
529 return processor
->grrr_runq
.count
;
533 sched_grrr_processor_runq_stats_count_sum(processor_t processor
)
535 return processor
->grrr_runq
.runq_stats
.count_sum
;
539 sched_grrr_processor_bound_count(__unused processor_t processor
)
545 sched_grrr_thread_update_scan(__unused sched_update_scan_context_t scan_context
)
550 #endif /* defined(CONFIG_SCHED_GRRR) */
552 #if defined(CONFIG_SCHED_GRRR_CORE)
555 grrr_priority_mapping_init(void)
559 /* Map 0->0 up to 10->20 */
560 for (i
= 0; i
<= 10; i
++) {
561 grrr_priority_mapping
[i
] = 2 * i
;
564 /* Map user priorities 11->33 up to 51 -> 153 */
565 for (i
= 11; i
<= 51; i
++) {
566 grrr_priority_mapping
[i
] = 3 * i
;
569 /* Map high priorities 52->180 up to 127->255 */
570 for (i
= 52; i
<= 127; i
++) {
571 grrr_priority_mapping
[i
] = 128 + i
;
574 for (i
= 0; i
< NUM_GRRR_PROPORTIONAL_PRIORITIES
; i
++) {
577 /* Calculate log(i); */
578 for (j
= 0, k
= 1; k
<= i
; j
++, k
*= 2) {
584 grrr_group_mapping
[i
] = i
>> 2;
589 grrr_intragroup_schedule(grrr_group_t group
)
593 if (group
->count
== 0) {
597 thread
= group
->current_client
;
598 if (thread
== THREAD_NULL
) {
599 thread
= (thread_t
)(void *)queue_first(&group
->clients
);
602 if (1 /* deficit */) {
603 group
->current_client
= (thread_t
)(void *)queue_next((queue_entry_t
)thread
);
604 if (queue_end(&group
->clients
, (queue_entry_t
)group
->current_client
)) {
605 group
->current_client
= (thread_t
)(void *)queue_first(&group
->clients
);
608 thread
= group
->current_client
;
615 grrr_intergroup_schedule(grrr_run_queue_t rq
)
620 if (rq
->count
== 0) {
624 group
= rq
->current_group
;
626 if (group
== GRRR_GROUP_NULL
) {
627 group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
630 thread
= grrr_intragroup_schedule(group
);
632 if ((group
->work
>= (UINT32_MAX
- 256)) || (rq
->last_rescale_tick
!= grrr_rescale_tick
)) {
633 grrr_rescale_work(rq
);
637 if (queue_end(&rq
->sorted_group_list
, queue_next((queue_entry_t
)group
))) {
638 /* last group, go back to beginning */
639 group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
641 grrr_group_t nextgroup
= (grrr_group_t
)queue_next((queue_entry_t
)group
);
642 uint64_t orderleft
, orderright
;
645 * The well-ordering condition for intergroup selection is:
647 * (group->work+1) / (nextgroup->work+1) > (group->weight) / (nextgroup->weight)
649 * Multiply both sides by their denominators to avoid division
652 orderleft
= (group
->work
+ 1) * ((uint64_t)nextgroup
->weight
);
653 orderright
= (nextgroup
->work
+ 1) * ((uint64_t)group
->weight
);
654 if (orderleft
> orderright
) {
657 group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
661 rq
->current_group
= group
;
667 grrr_runqueue_init(grrr_run_queue_t runq
)
669 grrr_group_index_t index
;
673 for (index
= 0; index
< NUM_GRRR_GROUPS
; index
++) {
674 unsigned int prisearch
;
677 prisearch
< NUM_GRRR_PROPORTIONAL_PRIORITIES
;
679 if (grrr_group_mapping
[prisearch
] == index
) {
680 runq
->groups
[index
].minpriority
= (grrr_proportional_priority_t
)prisearch
;
685 runq
->groups
[index
].index
= index
;
687 queue_init(&runq
->groups
[index
].clients
);
688 runq
->groups
[index
].count
= 0;
689 runq
->groups
[index
].weight
= 0;
690 runq
->groups
[index
].work
= 0;
691 runq
->groups
[index
].current_client
= THREAD_NULL
;
694 queue_init(&runq
->sorted_group_list
);
696 runq
->current_group
= GRRR_GROUP_NULL
;
700 grrr_rescale_work(grrr_run_queue_t rq
)
702 grrr_group_index_t index
;
704 /* avoid overflow by scaling by 1/8th */
705 for (index
= 0; index
< NUM_GRRR_GROUPS
; index
++) {
706 rq
->groups
[index
].work
>>= 3;
709 rq
->last_rescale_tick
= grrr_rescale_tick
;
717 grrr_proportional_priority_t gpriority
;
718 grrr_group_index_t gindex
;
721 gpriority
= grrr_priority_mapping
[thread
->sched_pri
];
722 gindex
= grrr_group_mapping
[gpriority
];
723 group
= &rq
->groups
[gindex
];
726 thread
->grrr_deficit
= 0;
729 if (group
->count
== 0) {
730 /* Empty group, this is the first client */
731 enqueue_tail(&group
->clients
, (queue_entry_t
)thread
);
733 group
->weight
= gpriority
;
734 group
->current_client
= thread
;
736 /* Insert before the current client */
737 if (group
->current_client
== THREAD_NULL
||
738 queue_first(&group
->clients
) == (queue_entry_t
)group
->current_client
) {
739 enqueue_head(&group
->clients
, (queue_entry_t
)thread
);
741 insque((queue_entry_t
)thread
, queue_prev((queue_entry_t
)group
->current_client
));
743 SCHED_STATS_RUNQ_CHANGE(&rq
->runq_stats
, rq
->count
);
745 group
->weight
+= gpriority
;
747 /* Since there was already a client, this is on the per-processor sorted list already */
748 remqueue((queue_entry_t
)group
);
751 grrr_sorted_list_insert_group(rq
, group
);
754 rq
->weight
+= gpriority
;
760 grrr_select(grrr_run_queue_t rq
)
764 thread
= grrr_intergroup_schedule(rq
);
765 if (thread
!= THREAD_NULL
) {
766 grrr_proportional_priority_t gpriority
;
767 grrr_group_index_t gindex
;
770 gpriority
= grrr_priority_mapping
[thread
->sched_pri
];
771 gindex
= grrr_group_mapping
[gpriority
];
772 group
= &rq
->groups
[gindex
];
774 remqueue((queue_entry_t
)thread
);
775 SCHED_STATS_RUNQ_CHANGE(&rq
->runq_stats
, rq
->count
);
777 group
->weight
-= gpriority
;
778 if (group
->current_client
== thread
) {
779 group
->current_client
= THREAD_NULL
;
782 remqueue((queue_entry_t
)group
);
783 if (group
->count
== 0) {
784 if (rq
->current_group
== group
) {
785 rq
->current_group
= GRRR_GROUP_NULL
;
788 /* Need to re-insert in sorted location */
789 grrr_sorted_list_insert_group(rq
, group
);
793 rq
->weight
-= gpriority
;
795 thread
->runq
= PROCESSOR_NULL
;
806 grrr_proportional_priority_t gpriority
;
807 grrr_group_index_t gindex
;
810 gpriority
= grrr_priority_mapping
[thread
->sched_pri
];
811 gindex
= grrr_group_mapping
[gpriority
];
812 group
= &rq
->groups
[gindex
];
814 remqueue((queue_entry_t
)thread
);
815 SCHED_STATS_RUNQ_CHANGE(&rq
->runq_stats
, rq
->count
);
817 group
->weight
-= gpriority
;
818 if (group
->current_client
== thread
) {
819 group
->current_client
= THREAD_NULL
;
822 remqueue((queue_entry_t
)group
);
823 if (group
->count
== 0) {
824 if (rq
->current_group
== group
) {
825 rq
->current_group
= GRRR_GROUP_NULL
;
828 /* Need to re-insert in sorted location */
829 grrr_sorted_list_insert_group(rq
, group
);
833 rq
->weight
-= gpriority
;
835 thread
->runq
= PROCESSOR_NULL
;
839 grrr_sorted_list_insert_group(grrr_run_queue_t rq
,
842 /* Simple insertion sort */
843 if (queue_empty(&rq
->sorted_group_list
)) {
844 enqueue_tail(&rq
->sorted_group_list
, (queue_entry_t
)group
);
846 grrr_group_t search_group
;
848 /* Start searching from the head (heaviest weight) for the first
849 * element less than us, so we can insert before it
851 search_group
= (grrr_group_t
)queue_first(&rq
->sorted_group_list
);
852 while (!queue_end(&rq
->sorted_group_list
, (queue_entry_t
)search_group
)) {
853 if (search_group
->weight
< group
->weight
) {
854 /* we should be before this */
855 search_group
= (grrr_group_t
)queue_prev((queue_entry_t
)search_group
);
858 if (search_group
->weight
== group
->weight
) {
859 /* Use group index as a tie breaker */
860 if (search_group
->index
< group
->index
) {
861 search_group
= (grrr_group_t
)queue_prev((queue_entry_t
)search_group
);
866 /* otherwise, our weight is too small, keep going */
867 search_group
= (grrr_group_t
)queue_next((queue_entry_t
)search_group
);
870 if (queue_end(&rq
->sorted_group_list
, (queue_entry_t
)search_group
)) {
871 enqueue_tail(&rq
->sorted_group_list
, (queue_entry_t
)group
);
873 insque((queue_entry_t
)group
, (queue_entry_t
)search_group
);
878 #endif /* defined(CONFIG_SCHED_GRRR_CORE) */