1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25#include <linux/cpu.h>
26#include <linux/cpumask.h>
27#include <linux/cpuset.h>
28#include <linux/err.h>
29#include <linux/errno.h>
30#include <linux/file.h>
31#include <linux/fs.h>
32#include <linux/init.h>
33#include <linux/interrupt.h>
34#include <linux/kernel.h>
35#include <linux/kmod.h>
36#include <linux/list.h>
37#include <linux/mempolicy.h>
38#include <linux/mm.h>
39#include <linux/memory.h>
40#include <linux/export.h>
41#include <linux/mount.h>
42#include <linux/namei.h>
43#include <linux/pagemap.h>
44#include <linux/proc_fs.h>
45#include <linux/rcupdate.h>
46#include <linux/sched.h>
47#include <linux/seq_file.h>
48#include <linux/security.h>
49#include <linux/slab.h>
50#include <linux/spinlock.h>
51#include <linux/stat.h>
52#include <linux/string.h>
53#include <linux/time.h>
54#include <linux/backing-dev.h>
55#include <linux/sort.h>
56
57#include <asm/uaccess.h>
58#include <linux/atomic.h>
59#include <linux/mutex.h>
60#include <linux/workqueue.h>
61#include <linux/cgroup.h>
62
63
64
65
66
67
68int number_of_cpusets __read_mostly;
69
70
71struct cgroup_subsys cpuset_subsys;
72struct cpuset;
73
74
75
76struct fmeter {
77 int cnt;
78 int val;
79 time_t time;
80 spinlock_t lock;
81};
82
83struct cpuset {
84 struct cgroup_subsys_state css;
85
86 unsigned long flags;
87 cpumask_var_t cpus_allowed;
88 nodemask_t mems_allowed;
89
90 struct fmeter fmeter;
91
92
93
94
95
96 int attach_in_progress;
97
98
99 int pn;
100
101
102 int relax_domain_level;
103
104 struct work_struct hotplug_work;
105};
106
107
108static inline struct cpuset *cgroup_cs(struct cgroup *cont)
109{
110 return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
111 struct cpuset, css);
112}
113
114
115static inline struct cpuset *task_cs(struct task_struct *task)
116{
117 return container_of(task_subsys_state(task, cpuset_subsys_id),
118 struct cpuset, css);
119}
120
121static inline struct cpuset *parent_cs(const struct cpuset *cs)
122{
123 struct cgroup *pcgrp = cs->css.cgroup->parent;
124
125 if (pcgrp)
126 return cgroup_cs(pcgrp);
127 return NULL;
128}
129
130#ifdef CONFIG_NUMA
131static inline bool task_has_mempolicy(struct task_struct *task)
132{
133 return task->mempolicy;
134}
135#else
136static inline bool task_has_mempolicy(struct task_struct *task)
137{
138 return false;
139}
140#endif
141
142
143
144typedef enum {
145 CS_ONLINE,
146 CS_CPU_EXCLUSIVE,
147 CS_MEM_EXCLUSIVE,
148 CS_MEM_HARDWALL,
149 CS_MEMORY_MIGRATE,
150 CS_SCHED_LOAD_BALANCE,
151 CS_SPREAD_PAGE,
152 CS_SPREAD_SLAB,
153} cpuset_flagbits_t;
154
155
156static inline bool is_cpuset_online(const struct cpuset *cs)
157{
158 return test_bit(CS_ONLINE, &cs->flags);
159}
160
161static inline int is_cpu_exclusive(const struct cpuset *cs)
162{
163 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
164}
165
166static inline int is_mem_exclusive(const struct cpuset *cs)
167{
168 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
169}
170
171static inline int is_mem_hardwall(const struct cpuset *cs)
172{
173 return test_bit(CS_MEM_HARDWALL, &cs->flags);
174}
175
176static inline int is_sched_load_balance(const struct cpuset *cs)
177{
178 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
179}
180
181static inline int is_memory_migrate(const struct cpuset *cs)
182{
183 return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
184}
185
186static inline int is_spread_page(const struct cpuset *cs)
187{
188 return test_bit(CS_SPREAD_PAGE, &cs->flags);
189}
190
191static inline int is_spread_slab(const struct cpuset *cs)
192{
193 return test_bit(CS_SPREAD_SLAB, &cs->flags);
194}
195
196static struct cpuset top_cpuset = {
197 .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
198 (1 << CS_MEM_EXCLUSIVE)),
199};
200
201
202
203
204
205
206
207
208
209
210#define cpuset_for_each_child(child_cs, pos_cgrp, parent_cs) \
211 cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup) \
212 if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp)))))
213
214
215
216
217
218
219
220
221
222
223
224#define cpuset_for_each_descendant_pre(des_cs, pos_cgrp, root_cs) \
225 cgroup_for_each_descendant_pre((pos_cgrp), (root_cs)->css.cgroup) \
226 if (is_cpuset_online(((des_cs) = cgroup_cs((pos_cgrp)))))
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264static DEFINE_MUTEX(cpuset_mutex);
265static DEFINE_MUTEX(callback_mutex);
266
267
268
269
270static struct workqueue_struct *cpuset_propagate_hotplug_wq;
271
272static void cpuset_hotplug_workfn(struct work_struct *work);
273static void cpuset_propagate_hotplug_workfn(struct work_struct *work);
274static void schedule_cpuset_propagate_hotplug(struct cpuset *cs);
275
276static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);
277
278
279
280
281
282
283static struct dentry *cpuset_mount(struct file_system_type *fs_type,
284 int flags, const char *unused_dev_name, void *data)
285{
286 struct file_system_type *cgroup_fs = get_fs_type("cgroup");
287 struct dentry *ret = ERR_PTR(-ENODEV);
288 if (cgroup_fs) {
289 char mountopts[] =
290 "cpuset,noprefix,"
291 "release_agent=/sbin/cpuset_release_agent";
292 ret = cgroup_fs->mount(cgroup_fs, flags,
293 unused_dev_name, mountopts);
294 put_filesystem(cgroup_fs);
295 }
296 return ret;
297}
298
299static struct file_system_type cpuset_fs_type = {
300 .name = "cpuset",
301 .mount = cpuset_mount,
302};
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318static void guarantee_online_cpus(const struct cpuset *cs,
319 struct cpumask *pmask)
320{
321 while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
322 cs = parent_cs(cs);
323 if (cs)
324 cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
325 else
326 cpumask_copy(pmask, cpu_online_mask);
327 BUG_ON(!cpumask_intersects(pmask, cpu_online_mask));
328}
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
344{
345 while (cs && !nodes_intersects(cs->mems_allowed,
346 node_states[N_MEMORY]))
347 cs = parent_cs(cs);
348 if (cs)
349 nodes_and(*pmask, cs->mems_allowed,
350 node_states[N_MEMORY]);
351 else
352 *pmask = node_states[N_MEMORY];
353 BUG_ON(!nodes_intersects(*pmask, node_states[N_MEMORY]));
354}
355
356
357
358
359
360
361static void cpuset_update_task_spread_flag(struct cpuset *cs,
362 struct task_struct *tsk)
363{
364 if (is_spread_page(cs))
365 tsk->flags |= PF_SPREAD_PAGE;
366 else
367 tsk->flags &= ~PF_SPREAD_PAGE;
368 if (is_spread_slab(cs))
369 tsk->flags |= PF_SPREAD_SLAB;
370 else
371 tsk->flags &= ~PF_SPREAD_SLAB;
372}
373
374
375
376
377
378
379
380
381
382static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
383{
384 return cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
385 nodes_subset(p->mems_allowed, q->mems_allowed) &&
386 is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
387 is_mem_exclusive(p) <= is_mem_exclusive(q);
388}
389
390
391
392
393
394static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs)
395{
396 struct cpuset *trial;
397
398 trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
399 if (!trial)
400 return NULL;
401
402 if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) {
403 kfree(trial);
404 return NULL;
405 }
406 cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
407
408 return trial;
409}
410
411
412
413
414
415static void free_trial_cpuset(struct cpuset *trial)
416{
417 free_cpumask_var(trial->cpus_allowed);
418 kfree(trial);
419}
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
442{
443 struct cgroup *cont;
444 struct cpuset *c, *par;
445 int ret;
446
447 rcu_read_lock();
448
449
450 ret = -EBUSY;
451 cpuset_for_each_child(c, cont, cur)
452 if (!is_cpuset_subset(c, trial))
453 goto out;
454
455
456 ret = 0;
457 if (cur == &top_cpuset)
458 goto out;
459
460 par = parent_cs(cur);
461
462
463 ret = -EACCES;
464 if (!is_cpuset_subset(trial, par))
465 goto out;
466
467
468
469
470
471 ret = -EINVAL;
472 cpuset_for_each_child(c, cont, par) {
473 if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
474 c != cur &&
475 cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
476 goto out;
477 if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
478 c != cur &&
479 nodes_intersects(trial->mems_allowed, c->mems_allowed))
480 goto out;
481 }
482
483
484
485
486
487 ret = -ENOSPC;
488 if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) &&
489 (cpumask_empty(trial->cpus_allowed) ||
490 nodes_empty(trial->mems_allowed)))
491 goto out;
492
493 ret = 0;
494out:
495 rcu_read_unlock();
496 return ret;
497}
498
499#ifdef CONFIG_SMP
500
501
502
503
504static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
505{
506 return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
507}
508
509static void
510update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
511{
512 if (dattr->relax_domain_level < c->relax_domain_level)
513 dattr->relax_domain_level = c->relax_domain_level;
514 return;
515}
516
517static void update_domain_attr_tree(struct sched_domain_attr *dattr,
518 struct cpuset *root_cs)
519{
520 struct cpuset *cp;
521 struct cgroup *pos_cgrp;
522
523 rcu_read_lock();
524 cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) {
525
526 if (cpumask_empty(cp->cpus_allowed)) {
527 pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
528 continue;
529 }
530
531 if (is_sched_load_balance(cp))
532 update_domain_attr(dattr, cp);
533 }
534 rcu_read_unlock();
535}
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591static int generate_sched_domains(cpumask_var_t **domains,
592 struct sched_domain_attr **attributes)
593{
594 struct cpuset *cp;
595 struct cpuset **csa;
596 int csn;
597 int i, j, k;
598 cpumask_var_t *doms;
599 struct sched_domain_attr *dattr;
600 int ndoms = 0;
601 int nslot;
602 struct cgroup *pos_cgrp;
603
604 doms = NULL;
605 dattr = NULL;
606 csa = NULL;
607
608
609 if (is_sched_load_balance(&top_cpuset)) {
610 ndoms = 1;
611 doms = alloc_sched_domains(ndoms);
612 if (!doms)
613 goto done;
614
615 dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
616 if (dattr) {
617 *dattr = SD_ATTR_INIT;
618 update_domain_attr_tree(dattr, &top_cpuset);
619 }
620 cpumask_copy(doms[0], top_cpuset.cpus_allowed);
621
622 goto done;
623 }
624
625 csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
626 if (!csa)
627 goto done;
628 csn = 0;
629
630 rcu_read_lock();
631 cpuset_for_each_descendant_pre(cp, pos_cgrp, &top_cpuset) {
632
633
634
635
636
637
638
639
640 if (!cpumask_empty(cp->cpus_allowed) &&
641 !is_sched_load_balance(cp))
642 continue;
643
644 if (is_sched_load_balance(cp))
645 csa[csn++] = cp;
646
647
648 pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
649 }
650 rcu_read_unlock();
651
652 for (i = 0; i < csn; i++)
653 csa[i]->pn = i;
654 ndoms = csn;
655
656restart:
657
658 for (i = 0; i < csn; i++) {
659 struct cpuset *a = csa[i];
660 int apn = a->pn;
661
662 for (j = 0; j < csn; j++) {
663 struct cpuset *b = csa[j];
664 int bpn = b->pn;
665
666 if (apn != bpn && cpusets_overlap(a, b)) {
667 for (k = 0; k < csn; k++) {
668 struct cpuset *c = csa[k];
669
670 if (c->pn == bpn)
671 c->pn = apn;
672 }
673 ndoms--;
674 goto restart;
675 }
676 }
677 }
678
679
680
681
682
683 doms = alloc_sched_domains(ndoms);
684 if (!doms)
685 goto done;
686
687
688
689
690
691 dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
692
693 for (nslot = 0, i = 0; i < csn; i++) {
694 struct cpuset *a = csa[i];
695 struct cpumask *dp;
696 int apn = a->pn;
697
698 if (apn < 0) {
699
700 continue;
701 }
702
703 dp = doms[nslot];
704
705 if (nslot == ndoms) {
706 static int warnings = 10;
707 if (warnings) {
708 printk(KERN_WARNING
709 "rebuild_sched_domains confused:"
710 " nslot %d, ndoms %d, csn %d, i %d,"
711 " apn %d\n",
712 nslot, ndoms, csn, i, apn);
713 warnings--;
714 }
715 continue;
716 }
717
718 cpumask_clear(dp);
719 if (dattr)
720 *(dattr + nslot) = SD_ATTR_INIT;
721 for (j = i; j < csn; j++) {
722 struct cpuset *b = csa[j];
723
724 if (apn == b->pn) {
725 cpumask_or(dp, dp, b->cpus_allowed);
726 if (dattr)
727 update_domain_attr_tree(dattr + nslot, b);
728
729
730 b->pn = -1;
731 }
732 }
733 nslot++;
734 }
735 BUG_ON(nslot != ndoms);
736
737done:
738 kfree(csa);
739
740
741
742
743
744 if (doms == NULL)
745 ndoms = 1;
746
747 *domains = doms;
748 *attributes = dattr;
749 return ndoms;
750}
751
752
753
754
755
756
757
758
759
760
761
762
763static void rebuild_sched_domains_locked(void)
764{
765 struct sched_domain_attr *attr;
766 cpumask_var_t *doms;
767 int ndoms;
768
769 lockdep_assert_held(&cpuset_mutex);
770 get_online_cpus();
771
772
773
774
775
776
777 if (!cpumask_equal(top_cpuset.cpus_allowed, cpu_active_mask))
778 goto out;
779
780
781 ndoms = generate_sched_domains(&doms, &attr);
782
783
784 partition_sched_domains(ndoms, doms, attr);
785out:
786 put_online_cpus();
787}
788#else
789static void rebuild_sched_domains_locked(void)
790{
791}
792#endif
793
794void rebuild_sched_domains(void)
795{
796 mutex_lock(&cpuset_mutex);
797 rebuild_sched_domains_locked();
798 mutex_unlock(&cpuset_mutex);
799}
800
801
802
803
804
805
806
807
808
809
810
811static int cpuset_test_cpumask(struct task_struct *tsk,
812 struct cgroup_scanner *scan)
813{
814 return !cpumask_equal(&tsk->cpus_allowed,
815 (cgroup_cs(scan->cg))->cpus_allowed);
816}
817
818
819
820
821
822
823
824
825
826
827
828
829static void cpuset_change_cpumask(struct task_struct *tsk,
830 struct cgroup_scanner *scan)
831{
832 set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
833}
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
849{
850 struct cgroup_scanner scan;
851
852 scan.cg = cs->css.cgroup;
853 scan.test_task = cpuset_test_cpumask;
854 scan.process_task = cpuset_change_cpumask;
855 scan.heap = heap;
856 cgroup_scan_tasks(&scan);
857}
858
859
860
861
862
863
864static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
865 const char *buf)
866{
867 struct ptr_heap heap;
868 int retval;
869 int is_load_balanced;
870
871
872 if (cs == &top_cpuset)
873 return -EACCES;
874
875
876
877
878
879
880
881 if (!*buf) {
882 cpumask_clear(trialcs->cpus_allowed);
883 } else {
884 retval = cpulist_parse(buf, trialcs->cpus_allowed);
885 if (retval < 0)
886 return retval;
887
888 if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
889 return -EINVAL;
890 }
891 retval = validate_change(cs, trialcs);
892 if (retval < 0)
893 return retval;
894
895
896 if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
897 return 0;
898
899 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
900 if (retval)
901 return retval;
902
903 is_load_balanced = is_sched_load_balance(trialcs);
904
905 mutex_lock(&callback_mutex);
906 cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
907 mutex_unlock(&callback_mutex);
908
909
910
911
912
913 update_tasks_cpumask(cs, &heap);
914
915 heap_free(&heap);
916
917 if (is_load_balanced)
918 rebuild_sched_domains_locked();
919 return 0;
920}
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
943 const nodemask_t *to)
944{
945 struct task_struct *tsk = current;
946
947 tsk->mems_allowed = *to;
948
949 do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
950
951 guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
952}
953
954
955
956
957
958
959
960
961
962
963static void cpuset_change_task_nodemask(struct task_struct *tsk,
964 nodemask_t *newmems)
965{
966 bool need_loop;
967
968
969
970
971
972 if (unlikely(test_thread_flag(TIF_MEMDIE)))
973 return;
974 if (current->flags & PF_EXITING)
975 return;
976
977 task_lock(tsk);
978
979
980
981
982
983
984 need_loop = task_has_mempolicy(tsk) ||
985 !nodes_intersects(*newmems, tsk->mems_allowed);
986
987 if (need_loop)
988 write_seqcount_begin(&tsk->mems_allowed_seq);
989
990 nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
991 mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
992
993 mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
994 tsk->mems_allowed = *newmems;
995
996 if (need_loop)
997 write_seqcount_end(&tsk->mems_allowed_seq);
998
999 task_unlock(tsk);
1000}
1001
1002
1003
1004
1005
1006
1007static void cpuset_change_nodemask(struct task_struct *p,
1008 struct cgroup_scanner *scan)
1009{
1010 struct mm_struct *mm;
1011 struct cpuset *cs;
1012 int migrate;
1013 const nodemask_t *oldmem = scan->data;
1014 static nodemask_t newmems;
1015
1016 cs = cgroup_cs(scan->cg);
1017 guarantee_online_mems(cs, &newmems);
1018
1019 cpuset_change_task_nodemask(p, &newmems);
1020
1021 mm = get_task_mm(p);
1022 if (!mm)
1023 return;
1024
1025 migrate = is_memory_migrate(cs);
1026
1027 mpol_rebind_mm(mm, &cs->mems_allowed);
1028 if (migrate)
1029 cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1030 mmput(mm);
1031}
1032
1033static void *cpuset_being_rebound;
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
1046 struct ptr_heap *heap)
1047{
1048 struct cgroup_scanner scan;
1049
1050 cpuset_being_rebound = cs;
1051
1052 scan.cg = cs->css.cgroup;
1053 scan.test_task = NULL;
1054 scan.process_task = cpuset_change_nodemask;
1055 scan.heap = heap;
1056 scan.data = (nodemask_t *)oldmem;
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068 cgroup_scan_tasks(&scan);
1069
1070
1071 cpuset_being_rebound = NULL;
1072}
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
1088 const char *buf)
1089{
1090 NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1091 int retval;
1092 struct ptr_heap heap;
1093
1094 if (!oldmem)
1095 return -ENOMEM;
1096
1097
1098
1099
1100
1101 if (cs == &top_cpuset) {
1102 retval = -EACCES;
1103 goto done;
1104 }
1105
1106
1107
1108
1109
1110
1111
1112 if (!*buf) {
1113 nodes_clear(trialcs->mems_allowed);
1114 } else {
1115 retval = nodelist_parse(buf, trialcs->mems_allowed);
1116 if (retval < 0)
1117 goto done;
1118
1119 if (!nodes_subset(trialcs->mems_allowed,
1120 node_states[N_MEMORY])) {
1121 retval = -EINVAL;
1122 goto done;
1123 }
1124 }
1125 *oldmem = cs->mems_allowed;
1126 if (nodes_equal(*oldmem, trialcs->mems_allowed)) {
1127 retval = 0;
1128 goto done;
1129 }
1130 retval = validate_change(cs, trialcs);
1131 if (retval < 0)
1132 goto done;
1133
1134 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1135 if (retval < 0)
1136 goto done;
1137
1138 mutex_lock(&callback_mutex);
1139 cs->mems_allowed = trialcs->mems_allowed;
1140 mutex_unlock(&callback_mutex);
1141
1142 update_tasks_nodemask(cs, oldmem, &heap);
1143
1144 heap_free(&heap);
1145done:
1146 NODEMASK_FREE(oldmem);
1147 return retval;
1148}
1149
1150int current_cpuset_is_being_rebound(void)
1151{
1152 return task_cs(current) == cpuset_being_rebound;
1153}
1154
1155static int update_relax_domain_level(struct cpuset *cs, s64 val)
1156{
1157#ifdef CONFIG_SMP
1158 if (val < -1 || val >= sched_domain_level_max)
1159 return -EINVAL;
1160#endif
1161
1162 if (val != cs->relax_domain_level) {
1163 cs->relax_domain_level = val;
1164 if (!cpumask_empty(cs->cpus_allowed) &&
1165 is_sched_load_balance(cs))
1166 rebuild_sched_domains_locked();
1167 }
1168
1169 return 0;
1170}
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182static void cpuset_change_flag(struct task_struct *tsk,
1183 struct cgroup_scanner *scan)
1184{
1185 cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk);
1186}
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap)
1202{
1203 struct cgroup_scanner scan;
1204
1205 scan.cg = cs->css.cgroup;
1206 scan.test_task = NULL;
1207 scan.process_task = cpuset_change_flag;
1208 scan.heap = heap;
1209 cgroup_scan_tasks(&scan);
1210}
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
1222 int turning_on)
1223{
1224 struct cpuset *trialcs;
1225 int balance_flag_changed;
1226 int spread_flag_changed;
1227 struct ptr_heap heap;
1228 int err;
1229
1230 trialcs = alloc_trial_cpuset(cs);
1231 if (!trialcs)
1232 return -ENOMEM;
1233
1234 if (turning_on)
1235 set_bit(bit, &trialcs->flags);
1236 else
1237 clear_bit(bit, &trialcs->flags);
1238
1239 err = validate_change(cs, trialcs);
1240 if (err < 0)
1241 goto out;
1242
1243 err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1244 if (err < 0)
1245 goto out;
1246
1247 balance_flag_changed = (is_sched_load_balance(cs) !=
1248 is_sched_load_balance(trialcs));
1249
1250 spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs))
1251 || (is_spread_page(cs) != is_spread_page(trialcs)));
1252
1253 mutex_lock(&callback_mutex);
1254 cs->flags = trialcs->flags;
1255 mutex_unlock(&callback_mutex);
1256
1257 if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed)
1258 rebuild_sched_domains_locked();
1259
1260 if (spread_flag_changed)
1261 update_tasks_flags(cs, &heap);
1262 heap_free(&heap);
1263out:
1264 free_trial_cpuset(trialcs);
1265 return err;
1266}
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313#define FM_COEF 933
1314#define FM_MAXTICKS ((time_t)99)
1315#define FM_MAXCNT 1000000
1316#define FM_SCALE 1000
1317
1318
1319static void fmeter_init(struct fmeter *fmp)
1320{
1321 fmp->cnt = 0;
1322 fmp->val = 0;
1323 fmp->time = 0;
1324 spin_lock_init(&fmp->lock);
1325}
1326
1327
1328static void fmeter_update(struct fmeter *fmp)
1329{
1330 time_t now = get_seconds();
1331 time_t ticks = now - fmp->time;
1332
1333 if (ticks == 0)
1334 return;
1335
1336 ticks = min(FM_MAXTICKS, ticks);
1337 while (ticks-- > 0)
1338 fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1339 fmp->time = now;
1340
1341 fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1342 fmp->cnt = 0;
1343}
1344
1345
1346static void fmeter_markevent(struct fmeter *fmp)
1347{
1348 spin_lock(&fmp->lock);
1349 fmeter_update(fmp);
1350 fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1351 spin_unlock(&fmp->lock);
1352}
1353
1354
1355static int fmeter_getrate(struct fmeter *fmp)
1356{
1357 int val;
1358
1359 spin_lock(&fmp->lock);
1360 fmeter_update(fmp);
1361 val = fmp->val;
1362 spin_unlock(&fmp->lock);
1363 return val;
1364}
1365
1366
1367static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1368{
1369 struct cpuset *cs = cgroup_cs(cgrp);
1370 struct task_struct *task;
1371 int ret;
1372
1373 mutex_lock(&cpuset_mutex);
1374
1375 ret = -ENOSPC;
1376 if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1377 goto out_unlock;
1378
1379 cgroup_taskset_for_each(task, cgrp, tset) {
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389 ret = -EINVAL;
1390 if (task->flags & PF_NO_SETAFFINITY)
1391 goto out_unlock;
1392 ret = security_task_setscheduler(task);
1393 if (ret)
1394 goto out_unlock;
1395 }
1396
1397
1398
1399
1400
1401 cs->attach_in_progress++;
1402 ret = 0;
1403out_unlock:
1404 mutex_unlock(&cpuset_mutex);
1405 return ret;
1406}
1407
1408static void cpuset_cancel_attach(struct cgroup *cgrp,
1409 struct cgroup_taskset *tset)
1410{
1411 mutex_lock(&cpuset_mutex);
1412 cgroup_cs(cgrp)->attach_in_progress--;
1413 mutex_unlock(&cpuset_mutex);
1414}
1415
1416
1417
1418
1419
1420
1421static cpumask_var_t cpus_attach;
1422
1423static void cpuset_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1424{
1425
1426 static nodemask_t cpuset_attach_nodemask_from;
1427 static nodemask_t cpuset_attach_nodemask_to;
1428 struct mm_struct *mm;
1429 struct task_struct *task;
1430 struct task_struct *leader = cgroup_taskset_first(tset);
1431 struct cgroup *oldcgrp = cgroup_taskset_cur_cgroup(tset);
1432 struct cpuset *cs = cgroup_cs(cgrp);
1433 struct cpuset *oldcs = cgroup_cs(oldcgrp);
1434
1435 mutex_lock(&cpuset_mutex);
1436
1437
1438 if (cs == &top_cpuset)
1439 cpumask_copy(cpus_attach, cpu_possible_mask);
1440 else
1441 guarantee_online_cpus(cs, cpus_attach);
1442
1443 guarantee_online_mems(cs, &cpuset_attach_nodemask_to);
1444
1445 cgroup_taskset_for_each(task, cgrp, tset) {
1446
1447
1448
1449
1450 WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach));
1451
1452 cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to);
1453 cpuset_update_task_spread_flag(cs, task);
1454 }
1455
1456
1457
1458
1459
1460 cpuset_attach_nodemask_from = oldcs->mems_allowed;
1461 cpuset_attach_nodemask_to = cs->mems_allowed;
1462 mm = get_task_mm(leader);
1463 if (mm) {
1464 mpol_rebind_mm(mm, &cpuset_attach_nodemask_to);
1465 if (is_memory_migrate(cs))
1466 cpuset_migrate_mm(mm, &cpuset_attach_nodemask_from,
1467 &cpuset_attach_nodemask_to);
1468 mmput(mm);
1469 }
1470
1471 cs->attach_in_progress--;
1472
1473
1474
1475
1476
1477
1478 if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1479 schedule_cpuset_propagate_hotplug(cs);
1480
1481 mutex_unlock(&cpuset_mutex);
1482}
1483
1484
1485
1486typedef enum {
1487 FILE_MEMORY_MIGRATE,
1488 FILE_CPULIST,
1489 FILE_MEMLIST,
1490 FILE_CPU_EXCLUSIVE,
1491 FILE_MEM_EXCLUSIVE,
1492 FILE_MEM_HARDWALL,
1493 FILE_SCHED_LOAD_BALANCE,
1494 FILE_SCHED_RELAX_DOMAIN_LEVEL,
1495 FILE_MEMORY_PRESSURE_ENABLED,
1496 FILE_MEMORY_PRESSURE,
1497 FILE_SPREAD_PAGE,
1498 FILE_SPREAD_SLAB,
1499} cpuset_filetype_t;
1500
1501static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1502{
1503 struct cpuset *cs = cgroup_cs(cgrp);
1504 cpuset_filetype_t type = cft->private;
1505 int retval = -ENODEV;
1506
1507 mutex_lock(&cpuset_mutex);
1508 if (!is_cpuset_online(cs))
1509 goto out_unlock;
1510
1511 switch (type) {
1512 case FILE_CPU_EXCLUSIVE:
1513 retval = update_flag(CS_CPU_EXCLUSIVE, cs, val);
1514 break;
1515 case FILE_MEM_EXCLUSIVE:
1516 retval = update_flag(CS_MEM_EXCLUSIVE, cs, val);
1517 break;
1518 case FILE_MEM_HARDWALL:
1519 retval = update_flag(CS_MEM_HARDWALL, cs, val);
1520 break;
1521 case FILE_SCHED_LOAD_BALANCE:
1522 retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val);
1523 break;
1524 case FILE_MEMORY_MIGRATE:
1525 retval = update_flag(CS_MEMORY_MIGRATE, cs, val);
1526 break;
1527 case FILE_MEMORY_PRESSURE_ENABLED:
1528 cpuset_memory_pressure_enabled = !!val;
1529 break;
1530 case FILE_MEMORY_PRESSURE:
1531 retval = -EACCES;
1532 break;
1533 case FILE_SPREAD_PAGE:
1534 retval = update_flag(CS_SPREAD_PAGE, cs, val);
1535 break;
1536 case FILE_SPREAD_SLAB:
1537 retval = update_flag(CS_SPREAD_SLAB, cs, val);
1538 break;
1539 default:
1540 retval = -EINVAL;
1541 break;
1542 }
1543out_unlock:
1544 mutex_unlock(&cpuset_mutex);
1545 return retval;
1546}
1547
1548static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val)
1549{
1550 struct cpuset *cs = cgroup_cs(cgrp);
1551 cpuset_filetype_t type = cft->private;
1552 int retval = -ENODEV;
1553
1554 mutex_lock(&cpuset_mutex);
1555 if (!is_cpuset_online(cs))
1556 goto out_unlock;
1557
1558 switch (type) {
1559 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1560 retval = update_relax_domain_level(cs, val);
1561 break;
1562 default:
1563 retval = -EINVAL;
1564 break;
1565 }
1566out_unlock:
1567 mutex_unlock(&cpuset_mutex);
1568 return retval;
1569}
1570
1571
1572
1573
1574static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft,
1575 const char *buf)
1576{
1577 struct cpuset *cs = cgroup_cs(cgrp);
1578 struct cpuset *trialcs;
1579 int retval = -ENODEV;
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596 flush_work(&cpuset_hotplug_work);
1597 flush_workqueue(cpuset_propagate_hotplug_wq);
1598
1599 mutex_lock(&cpuset_mutex);
1600 if (!is_cpuset_online(cs))
1601 goto out_unlock;
1602
1603 trialcs = alloc_trial_cpuset(cs);
1604 if (!trialcs) {
1605 retval = -ENOMEM;
1606 goto out_unlock;
1607 }
1608
1609 switch (cft->private) {
1610 case FILE_CPULIST:
1611 retval = update_cpumask(cs, trialcs, buf);
1612 break;
1613 case FILE_MEMLIST:
1614 retval = update_nodemask(cs, trialcs, buf);
1615 break;
1616 default:
1617 retval = -EINVAL;
1618 break;
1619 }
1620
1621 free_trial_cpuset(trialcs);
1622out_unlock:
1623 mutex_unlock(&cpuset_mutex);
1624 return retval;
1625}
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639static size_t cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1640{
1641 size_t count;
1642
1643 mutex_lock(&callback_mutex);
1644 count = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed);
1645 mutex_unlock(&callback_mutex);
1646
1647 return count;
1648}
1649
1650static size_t cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1651{
1652 size_t count;
1653
1654 mutex_lock(&callback_mutex);
1655 count = nodelist_scnprintf(page, PAGE_SIZE, cs->mems_allowed);
1656 mutex_unlock(&callback_mutex);
1657
1658 return count;
1659}
1660
1661static ssize_t cpuset_common_file_read(struct cgroup *cont,
1662 struct cftype *cft,
1663 struct file *file,
1664 char __user *buf,
1665 size_t nbytes, loff_t *ppos)
1666{
1667 struct cpuset *cs = cgroup_cs(cont);
1668 cpuset_filetype_t type = cft->private;
1669 char *page;
1670 ssize_t retval = 0;
1671 char *s;
1672
1673 if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1674 return -ENOMEM;
1675
1676 s = page;
1677
1678 switch (type) {
1679 case FILE_CPULIST:
1680 s += cpuset_sprintf_cpulist(s, cs);
1681 break;
1682 case FILE_MEMLIST:
1683 s += cpuset_sprintf_memlist(s, cs);
1684 break;
1685 default:
1686 retval = -EINVAL;
1687 goto out;
1688 }
1689 *s++ = '\n';
1690
1691 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1692out:
1693 free_page((unsigned long)page);
1694 return retval;
1695}
1696
1697static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft)
1698{
1699 struct cpuset *cs = cgroup_cs(cont);
1700 cpuset_filetype_t type = cft->private;
1701 switch (type) {
1702 case FILE_CPU_EXCLUSIVE:
1703 return is_cpu_exclusive(cs);
1704 case FILE_MEM_EXCLUSIVE:
1705 return is_mem_exclusive(cs);
1706 case FILE_MEM_HARDWALL:
1707 return is_mem_hardwall(cs);
1708 case FILE_SCHED_LOAD_BALANCE:
1709 return is_sched_load_balance(cs);
1710 case FILE_MEMORY_MIGRATE:
1711 return is_memory_migrate(cs);
1712 case FILE_MEMORY_PRESSURE_ENABLED:
1713 return cpuset_memory_pressure_enabled;
1714 case FILE_MEMORY_PRESSURE:
1715 return fmeter_getrate(&cs->fmeter);
1716 case FILE_SPREAD_PAGE:
1717 return is_spread_page(cs);
1718 case FILE_SPREAD_SLAB:
1719 return is_spread_slab(cs);
1720 default:
1721 BUG();
1722 }
1723
1724
1725 return 0;
1726}
1727
1728static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft)
1729{
1730 struct cpuset *cs = cgroup_cs(cont);
1731 cpuset_filetype_t type = cft->private;
1732 switch (type) {
1733 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1734 return cs->relax_domain_level;
1735 default:
1736 BUG();
1737 }
1738
1739
1740 return 0;
1741}
1742
1743
1744
1745
1746
1747
1748static struct cftype files[] = {
1749 {
1750 .name = "cpus",
1751 .read = cpuset_common_file_read,
1752 .write_string = cpuset_write_resmask,
1753 .max_write_len = (100U + 6 * NR_CPUS),
1754 .private = FILE_CPULIST,
1755 },
1756
1757 {
1758 .name = "mems",
1759 .read = cpuset_common_file_read,
1760 .write_string = cpuset_write_resmask,
1761 .max_write_len = (100U + 6 * MAX_NUMNODES),
1762 .private = FILE_MEMLIST,
1763 },
1764
1765 {
1766 .name = "cpu_exclusive",
1767 .read_u64 = cpuset_read_u64,
1768 .write_u64 = cpuset_write_u64,
1769 .private = FILE_CPU_EXCLUSIVE,
1770 },
1771
1772 {
1773 .name = "mem_exclusive",
1774 .read_u64 = cpuset_read_u64,
1775 .write_u64 = cpuset_write_u64,
1776 .private = FILE_MEM_EXCLUSIVE,
1777 },
1778
1779 {
1780 .name = "mem_hardwall",
1781 .read_u64 = cpuset_read_u64,
1782 .write_u64 = cpuset_write_u64,
1783 .private = FILE_MEM_HARDWALL,
1784 },
1785
1786 {
1787 .name = "sched_load_balance",
1788 .read_u64 = cpuset_read_u64,
1789 .write_u64 = cpuset_write_u64,
1790 .private = FILE_SCHED_LOAD_BALANCE,
1791 },
1792
1793 {
1794 .name = "sched_relax_domain_level",
1795 .read_s64 = cpuset_read_s64,
1796 .write_s64 = cpuset_write_s64,
1797 .private = FILE_SCHED_RELAX_DOMAIN_LEVEL,
1798 },
1799
1800 {
1801 .name = "memory_migrate",
1802 .read_u64 = cpuset_read_u64,
1803 .write_u64 = cpuset_write_u64,
1804 .private = FILE_MEMORY_MIGRATE,
1805 },
1806
1807 {
1808 .name = "memory_pressure",
1809 .read_u64 = cpuset_read_u64,
1810 .write_u64 = cpuset_write_u64,
1811 .private = FILE_MEMORY_PRESSURE,
1812 .mode = S_IRUGO,
1813 },
1814
1815 {
1816 .name = "memory_spread_page",
1817 .read_u64 = cpuset_read_u64,
1818 .write_u64 = cpuset_write_u64,
1819 .private = FILE_SPREAD_PAGE,
1820 },
1821
1822 {
1823 .name = "memory_spread_slab",
1824 .read_u64 = cpuset_read_u64,
1825 .write_u64 = cpuset_write_u64,
1826 .private = FILE_SPREAD_SLAB,
1827 },
1828
1829 {
1830 .name = "memory_pressure_enabled",
1831 .flags = CFTYPE_ONLY_ON_ROOT,
1832 .read_u64 = cpuset_read_u64,
1833 .write_u64 = cpuset_write_u64,
1834 .private = FILE_MEMORY_PRESSURE_ENABLED,
1835 },
1836
1837 { }
1838};
1839
1840
1841
1842
1843
1844
1845static struct cgroup_subsys_state *cpuset_css_alloc(struct cgroup *cont)
1846{
1847 struct cpuset *cs;
1848
1849 if (!cont->parent)
1850 return &top_cpuset.css;
1851
1852 cs = kzalloc(sizeof(*cs), GFP_KERNEL);
1853 if (!cs)
1854 return ERR_PTR(-ENOMEM);
1855 if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) {
1856 kfree(cs);
1857 return ERR_PTR(-ENOMEM);
1858 }
1859
1860 set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
1861 cpumask_clear(cs->cpus_allowed);
1862 nodes_clear(cs->mems_allowed);
1863 fmeter_init(&cs->fmeter);
1864 INIT_WORK(&cs->hotplug_work, cpuset_propagate_hotplug_workfn);
1865 cs->relax_domain_level = -1;
1866
1867 return &cs->css;
1868}
1869
1870static int cpuset_css_online(struct cgroup *cgrp)
1871{
1872 struct cpuset *cs = cgroup_cs(cgrp);
1873 struct cpuset *parent = parent_cs(cs);
1874 struct cpuset *tmp_cs;
1875 struct cgroup *pos_cg;
1876
1877 if (!parent)
1878 return 0;
1879
1880 mutex_lock(&cpuset_mutex);
1881
1882 set_bit(CS_ONLINE, &cs->flags);
1883 if (is_spread_page(parent))
1884 set_bit(CS_SPREAD_PAGE, &cs->flags);
1885 if (is_spread_slab(parent))
1886 set_bit(CS_SPREAD_SLAB, &cs->flags);
1887
1888 number_of_cpusets++;
1889
1890 if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags))
1891 goto out_unlock;
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906 rcu_read_lock();
1907 cpuset_for_each_child(tmp_cs, pos_cg, parent) {
1908 if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) {
1909 rcu_read_unlock();
1910 goto out_unlock;
1911 }
1912 }
1913 rcu_read_unlock();
1914
1915 mutex_lock(&callback_mutex);
1916 cs->mems_allowed = parent->mems_allowed;
1917 cpumask_copy(cs->cpus_allowed, parent->cpus_allowed);
1918 mutex_unlock(&callback_mutex);
1919out_unlock:
1920 mutex_unlock(&cpuset_mutex);
1921 return 0;
1922}
1923
1924static void cpuset_css_offline(struct cgroup *cgrp)
1925{
1926 struct cpuset *cs = cgroup_cs(cgrp);
1927
1928 mutex_lock(&cpuset_mutex);
1929
1930 if (is_sched_load_balance(cs))
1931 update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
1932
1933 number_of_cpusets--;
1934 clear_bit(CS_ONLINE, &cs->flags);
1935
1936 mutex_unlock(&cpuset_mutex);
1937}
1938
1939
1940
1941
1942
1943
1944
1945static void cpuset_css_free(struct cgroup *cont)
1946{
1947 struct cpuset *cs = cgroup_cs(cont);
1948
1949 free_cpumask_var(cs->cpus_allowed);
1950 kfree(cs);
1951}
1952
1953struct cgroup_subsys cpuset_subsys = {
1954 .name = "cpuset",
1955 .css_alloc = cpuset_css_alloc,
1956 .css_online = cpuset_css_online,
1957 .css_offline = cpuset_css_offline,
1958 .css_free = cpuset_css_free,
1959 .can_attach = cpuset_can_attach,
1960 .cancel_attach = cpuset_cancel_attach,
1961 .attach = cpuset_attach,
1962 .subsys_id = cpuset_subsys_id,
1963 .base_cftypes = files,
1964 .early_init = 1,
1965};
1966
1967
1968
1969
1970
1971
1972
1973int __init cpuset_init(void)
1974{
1975 int err = 0;
1976
1977 if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL))
1978 BUG();
1979
1980 cpumask_setall(top_cpuset.cpus_allowed);
1981 nodes_setall(top_cpuset.mems_allowed);
1982
1983 fmeter_init(&top_cpuset.fmeter);
1984 set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1985 top_cpuset.relax_domain_level = -1;
1986
1987 err = register_filesystem(&cpuset_fs_type);
1988 if (err < 0)
1989 return err;
1990
1991 if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL))
1992 BUG();
1993
1994 number_of_cpusets = 1;
1995 return 0;
1996}
1997
1998
1999
2000
2001
2002
2003
2004
2005static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
2006{
2007 struct cpuset *parent;
2008
2009
2010
2011
2012
2013 parent = parent_cs(cs);
2014 while (cpumask_empty(parent->cpus_allowed) ||
2015 nodes_empty(parent->mems_allowed))
2016 parent = parent_cs(parent);
2017
2018 if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) {
2019 rcu_read_lock();
2020 printk(KERN_ERR "cpuset: failed to transfer tasks out of empty cpuset %s\n",
2021 cgroup_name(cs->css.cgroup));
2022 rcu_read_unlock();
2023 }
2024}
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034static void cpuset_propagate_hotplug_workfn(struct work_struct *work)
2035{
2036 static cpumask_t off_cpus;
2037 static nodemask_t off_mems, tmp_mems;
2038 struct cpuset *cs = container_of(work, struct cpuset, hotplug_work);
2039 bool is_empty;
2040
2041 mutex_lock(&cpuset_mutex);
2042
2043 cpumask_andnot(&off_cpus, cs->cpus_allowed, top_cpuset.cpus_allowed);
2044 nodes_andnot(off_mems, cs->mems_allowed, top_cpuset.mems_allowed);
2045
2046
2047 if (!cpumask_empty(&off_cpus)) {
2048 mutex_lock(&callback_mutex);
2049 cpumask_andnot(cs->cpus_allowed, cs->cpus_allowed, &off_cpus);
2050 mutex_unlock(&callback_mutex);
2051 update_tasks_cpumask(cs, NULL);
2052 }
2053
2054
2055 if (!nodes_empty(off_mems)) {
2056 tmp_mems = cs->mems_allowed;
2057 mutex_lock(&callback_mutex);
2058 nodes_andnot(cs->mems_allowed, cs->mems_allowed, off_mems);
2059 mutex_unlock(&callback_mutex);
2060 update_tasks_nodemask(cs, &tmp_mems, NULL);
2061 }
2062
2063 is_empty = cpumask_empty(cs->cpus_allowed) ||
2064 nodes_empty(cs->mems_allowed);
2065
2066 mutex_unlock(&cpuset_mutex);
2067
2068
2069
2070
2071
2072
2073 if (is_empty)
2074 remove_tasks_in_empty_cpuset(cs);
2075
2076
2077 css_put(&cs->css);
2078}
2079
2080
2081
2082
2083
2084
2085
2086
2087static void schedule_cpuset_propagate_hotplug(struct cpuset *cs)
2088{
2089
2090
2091
2092
2093 if (!css_tryget(&cs->css))
2094 return;
2095
2096
2097
2098
2099
2100
2101 if (!queue_work(cpuset_propagate_hotplug_wq, &cs->hotplug_work))
2102 css_put(&cs->css);
2103}
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121static void cpuset_hotplug_workfn(struct work_struct *work)
2122{
2123 static cpumask_t new_cpus, tmp_cpus;
2124 static nodemask_t new_mems, tmp_mems;
2125 bool cpus_updated, mems_updated;
2126 bool cpus_offlined, mems_offlined;
2127
2128 mutex_lock(&cpuset_mutex);
2129
2130
2131 cpumask_copy(&new_cpus, cpu_active_mask);
2132 new_mems = node_states[N_MEMORY];
2133
2134 cpus_updated = !cpumask_equal(top_cpuset.cpus_allowed, &new_cpus);
2135 cpus_offlined = cpumask_andnot(&tmp_cpus, top_cpuset.cpus_allowed,
2136 &new_cpus);
2137
2138 mems_updated = !nodes_equal(top_cpuset.mems_allowed, new_mems);
2139 nodes_andnot(tmp_mems, top_cpuset.mems_allowed, new_mems);
2140 mems_offlined = !nodes_empty(tmp_mems);
2141
2142
2143 if (cpus_updated) {
2144 mutex_lock(&callback_mutex);
2145 cpumask_copy(top_cpuset.cpus_allowed, &new_cpus);
2146 mutex_unlock(&callback_mutex);
2147
2148 }
2149
2150
2151 if (mems_updated) {
2152 tmp_mems = top_cpuset.mems_allowed;
2153 mutex_lock(&callback_mutex);
2154 top_cpuset.mems_allowed = new_mems;
2155 mutex_unlock(&callback_mutex);
2156 update_tasks_nodemask(&top_cpuset, &tmp_mems, NULL);
2157 }
2158
2159
2160 if (cpus_offlined || mems_offlined) {
2161 struct cpuset *cs;
2162 struct cgroup *pos_cgrp;
2163
2164 rcu_read_lock();
2165 cpuset_for_each_descendant_pre(cs, pos_cgrp, &top_cpuset)
2166 schedule_cpuset_propagate_hotplug(cs);
2167 rcu_read_unlock();
2168 }
2169
2170 mutex_unlock(&cpuset_mutex);
2171
2172
2173 flush_workqueue(cpuset_propagate_hotplug_wq);
2174
2175
2176 if (cpus_updated)
2177 rebuild_sched_domains();
2178}
2179
2180void cpuset_update_active_cpus(bool cpu_online)
2181{
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192 partition_sched_domains(1, NULL, NULL);
2193 schedule_work(&cpuset_hotplug_work);
2194}
2195
2196
2197
2198
2199
2200
2201static int cpuset_track_online_nodes(struct notifier_block *self,
2202 unsigned long action, void *arg)
2203{
2204 schedule_work(&cpuset_hotplug_work);
2205 return NOTIFY_OK;
2206}
2207
2208static struct notifier_block cpuset_track_online_nodes_nb = {
2209 .notifier_call = cpuset_track_online_nodes,
2210 .priority = 10,
2211};
2212
2213
2214
2215
2216
2217
2218void __init cpuset_init_smp(void)
2219{
2220 cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask);
2221 top_cpuset.mems_allowed = node_states[N_MEMORY];
2222
2223 register_hotmemory_notifier(&cpuset_track_online_nodes_nb);
2224
2225 cpuset_propagate_hotplug_wq =
2226 alloc_ordered_workqueue("cpuset_hotplug", 0);
2227 BUG_ON(!cpuset_propagate_hotplug_wq);
2228}
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
2242{
2243 mutex_lock(&callback_mutex);
2244 task_lock(tsk);
2245 guarantee_online_cpus(task_cs(tsk), pmask);
2246 task_unlock(tsk);
2247 mutex_unlock(&callback_mutex);
2248}
2249
2250void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
2251{
2252 const struct cpuset *cs;
2253
2254 rcu_read_lock();
2255 cs = task_cs(tsk);
2256 if (cs)
2257 do_set_cpus_allowed(tsk, cs->cpus_allowed);
2258 rcu_read_unlock();
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277}
2278
2279void cpuset_init_current_mems_allowed(void)
2280{
2281 nodes_setall(current->mems_allowed);
2282}
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
2295{
2296 nodemask_t mask;
2297
2298 mutex_lock(&callback_mutex);
2299 task_lock(tsk);
2300 guarantee_online_mems(task_cs(tsk), &mask);
2301 task_unlock(tsk);
2302 mutex_unlock(&callback_mutex);
2303
2304 return mask;
2305}
2306
2307
2308
2309
2310
2311
2312
2313int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
2314{
2315 return nodes_intersects(*nodemask, current->mems_allowed);
2316}
2317
2318
2319
2320
2321
2322
2323
2324static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2325{
2326 while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs))
2327 cs = parent_cs(cs);
2328 return cs;
2329}
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask)
2393{
2394 const struct cpuset *cs;
2395 int allowed;
2396
2397 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2398 return 1;
2399 might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2400 if (node_isset(node, current->mems_allowed))
2401 return 1;
2402
2403
2404
2405
2406 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2407 return 1;
2408 if (gfp_mask & __GFP_HARDWALL)
2409 return 0;
2410
2411 if (current->flags & PF_EXITING)
2412 return 1;
2413
2414
2415 mutex_lock(&callback_mutex);
2416
2417 task_lock(current);
2418 cs = nearest_hardwall_ancestor(task_cs(current));
2419 task_unlock(current);
2420
2421 allowed = node_isset(node, cs->mems_allowed);
2422 mutex_unlock(&callback_mutex);
2423 return allowed;
2424}
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask)
2450{
2451 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2452 return 1;
2453 if (node_isset(node, current->mems_allowed))
2454 return 1;
2455
2456
2457
2458
2459 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2460 return 1;
2461 return 0;
2462}
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491static int cpuset_spread_node(int *rotor)
2492{
2493 int node;
2494
2495 node = next_node(*rotor, current->mems_allowed);
2496 if (node == MAX_NUMNODES)
2497 node = first_node(current->mems_allowed);
2498 *rotor = node;
2499 return node;
2500}
2501
2502int cpuset_mem_spread_node(void)
2503{
2504 if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE)
2505 current->cpuset_mem_spread_rotor =
2506 node_random(¤t->mems_allowed);
2507
2508 return cpuset_spread_node(¤t->cpuset_mem_spread_rotor);
2509}
2510
2511int cpuset_slab_spread_node(void)
2512{
2513 if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE)
2514 current->cpuset_slab_spread_rotor =
2515 node_random(¤t->mems_allowed);
2516
2517 return cpuset_spread_node(¤t->cpuset_slab_spread_rotor);
2518}
2519
2520EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2534 const struct task_struct *tsk2)
2535{
2536 return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
2537}
2538
2539#define CPUSET_NODELIST_LEN (256)
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549void cpuset_print_task_mems_allowed(struct task_struct *tsk)
2550{
2551
2552 static char cpuset_nodelist[CPUSET_NODELIST_LEN];
2553 static DEFINE_SPINLOCK(cpuset_buffer_lock);
2554
2555 struct cgroup *cgrp = task_cs(tsk)->css.cgroup;
2556
2557 rcu_read_lock();
2558 spin_lock(&cpuset_buffer_lock);
2559
2560 nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN,
2561 tsk->mems_allowed);
2562 printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n",
2563 tsk->comm, cgroup_name(cgrp), cpuset_nodelist);
2564
2565 spin_unlock(&cpuset_buffer_lock);
2566 rcu_read_unlock();
2567}
2568
2569
2570
2571
2572
2573
2574
2575int cpuset_memory_pressure_enabled __read_mostly;
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595void __cpuset_memory_pressure_bump(void)
2596{
2597 task_lock(current);
2598 fmeter_markevent(&task_cs(current)->fmeter);
2599 task_unlock(current);
2600}
2601
2602#ifdef CONFIG_PROC_PID_CPUSET
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612int proc_cpuset_show(struct seq_file *m, void *unused_v)
2613{
2614 struct pid *pid;
2615 struct task_struct *tsk;
2616 char *buf;
2617 struct cgroup_subsys_state *css;
2618 int retval;
2619
2620 retval = -ENOMEM;
2621 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2622 if (!buf)
2623 goto out;
2624
2625 retval = -ESRCH;
2626 pid = m->private;
2627 tsk = get_pid_task(pid, PIDTYPE_PID);
2628 if (!tsk)
2629 goto out_free;
2630
2631 rcu_read_lock();
2632 css = task_subsys_state(tsk, cpuset_subsys_id);
2633 retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
2634 rcu_read_unlock();
2635 if (retval < 0)
2636 goto out_put_task;
2637 seq_puts(m, buf);
2638 seq_putc(m, '\n');
2639out_put_task:
2640 put_task_struct(tsk);
2641out_free:
2642 kfree(buf);
2643out:
2644 return retval;
2645}
2646#endif
2647
2648
2649void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task)
2650{
2651 seq_printf(m, "Mems_allowed:\t");
2652 seq_nodemask(m, &task->mems_allowed);
2653 seq_printf(m, "\n");
2654 seq_printf(m, "Mems_allowed_list:\t");
2655 seq_nodemask_list(m, &task->mems_allowed);
2656 seq_printf(m, "\n");
2657}
2658