aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/sched.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/sched.c')
-rw-r--r--kernel/sched.c1148
1 files changed, 597 insertions, 551 deletions
diff --git a/kernel/sched.c b/kernel/sched.c
index d2d16d1273b1..b309027bf9e8 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -118,6 +118,12 @@
118 */ 118 */
119#define RUNTIME_INF ((u64)~0ULL) 119#define RUNTIME_INF ((u64)~0ULL)
120 120
121DEFINE_TRACE(sched_wait_task);
122DEFINE_TRACE(sched_wakeup);
123DEFINE_TRACE(sched_wakeup_new);
124DEFINE_TRACE(sched_switch);
125DEFINE_TRACE(sched_migrate_task);
126
121#ifdef CONFIG_SMP 127#ifdef CONFIG_SMP
122/* 128/*
123 * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) 129 * Divide a load by a sched group cpu_power : (load / sg->__cpu_power)
@@ -261,6 +267,10 @@ struct task_group {
261 struct cgroup_subsys_state css; 267 struct cgroup_subsys_state css;
262#endif 268#endif
263 269
270#ifdef CONFIG_USER_SCHED
271 uid_t uid;
272#endif
273
264#ifdef CONFIG_FAIR_GROUP_SCHED 274#ifdef CONFIG_FAIR_GROUP_SCHED
265 /* schedulable entities of this group on each cpu */ 275 /* schedulable entities of this group on each cpu */
266 struct sched_entity **se; 276 struct sched_entity **se;
@@ -286,6 +296,12 @@ struct task_group {
286 296
287#ifdef CONFIG_USER_SCHED 297#ifdef CONFIG_USER_SCHED
288 298
299/* Helper function to pass uid information to create_sched_user() */
300void set_tg_uid(struct user_struct *user)
301{
302 user->tg->uid = user->uid;
303}
304
289/* 305/*
290 * Root task group. 306 * Root task group.
291 * Every UID task group (including init_task_group aka UID-0) will 307 * Every UID task group (including init_task_group aka UID-0) will
@@ -481,14 +497,14 @@ struct rt_rq {
481 */ 497 */
482struct root_domain { 498struct root_domain {
483 atomic_t refcount; 499 atomic_t refcount;
484 cpumask_t span; 500 cpumask_var_t span;
485 cpumask_t online; 501 cpumask_var_t online;
486 502
487 /* 503 /*
488 * The "RT overload" flag: it gets set if a CPU has more than 504 * The "RT overload" flag: it gets set if a CPU has more than
489 * one runnable RT task. 505 * one runnable RT task.
490 */ 506 */
491 cpumask_t rto_mask; 507 cpumask_var_t rto_mask;
492 atomic_t rto_count; 508 atomic_t rto_count;
493#ifdef CONFIG_SMP 509#ifdef CONFIG_SMP
494 struct cpupri cpupri; 510 struct cpupri cpupri;
@@ -703,45 +719,18 @@ static __read_mostly char *sched_feat_names[] = {
703 719
704#undef SCHED_FEAT 720#undef SCHED_FEAT
705 721
706static int sched_feat_open(struct inode *inode, struct file *filp) 722static int sched_feat_show(struct seq_file *m, void *v)
707{ 723{
708 filp->private_data = inode->i_private;
709 return 0;
710}
711
712static ssize_t
713sched_feat_read(struct file *filp, char __user *ubuf,
714 size_t cnt, loff_t *ppos)
715{
716 char *buf;
717 int r = 0;
718 int len = 0;
719 int i; 724 int i;
720 725
721 for (i = 0; sched_feat_names[i]; i++) { 726 for (i = 0; sched_feat_names[i]; i++) {
722 len += strlen(sched_feat_names[i]); 727 if (!(sysctl_sched_features & (1UL << i)))
723 len += 4; 728 seq_puts(m, "NO_");
724 } 729 seq_printf(m, "%s ", sched_feat_names[i]);
725
726 buf = kmalloc(len + 2, GFP_KERNEL);
727 if (!buf)
728 return -ENOMEM;
729
730 for (i = 0; sched_feat_names[i]; i++) {
731 if (sysctl_sched_features & (1UL << i))
732 r += sprintf(buf + r, "%s ", sched_feat_names[i]);
733 else
734 r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]);
735 } 730 }
731 seq_puts(m, "\n");
736 732
737 r += sprintf(buf + r, "\n"); 733 return 0;
738 WARN_ON(r >= len + 2);
739
740 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
741
742 kfree(buf);
743
744 return r;
745} 734}
746 735
747static ssize_t 736static ssize_t
@@ -786,10 +775,17 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
786 return cnt; 775 return cnt;
787} 776}
788 777
778static int sched_feat_open(struct inode *inode, struct file *filp)
779{
780 return single_open(filp, sched_feat_show, NULL);
781}
782
789static struct file_operations sched_feat_fops = { 783static struct file_operations sched_feat_fops = {
790 .open = sched_feat_open, 784 .open = sched_feat_open,
791 .read = sched_feat_read, 785 .write = sched_feat_write,
792 .write = sched_feat_write, 786 .read = seq_read,
787 .llseek = seq_lseek,
788 .release = single_release,
793}; 789};
794 790
795static __init int sched_init_debug(void) 791static __init int sched_init_debug(void)
@@ -1474,27 +1470,13 @@ static void
1474update_group_shares_cpu(struct task_group *tg, int cpu, 1470update_group_shares_cpu(struct task_group *tg, int cpu,
1475 unsigned long sd_shares, unsigned long sd_rq_weight) 1471 unsigned long sd_shares, unsigned long sd_rq_weight)
1476{ 1472{
1477 int boost = 0;
1478 unsigned long shares; 1473 unsigned long shares;
1479 unsigned long rq_weight; 1474 unsigned long rq_weight;
1480 1475
1481 if (!tg->se[cpu]) 1476 if (!tg->se[cpu])
1482 return; 1477 return;
1483 1478
1484 rq_weight = tg->cfs_rq[cpu]->load.weight; 1479 rq_weight = tg->cfs_rq[cpu]->rq_weight;
1485
1486 /*
1487 * If there are currently no tasks on the cpu pretend there is one of
1488 * average load so that when a new task gets to run here it will not
1489 * get delayed by group starvation.
1490 */
1491 if (!rq_weight) {
1492 boost = 1;
1493 rq_weight = NICE_0_LOAD;
1494 }
1495
1496 if (unlikely(rq_weight > sd_rq_weight))
1497 rq_weight = sd_rq_weight;
1498 1480
1499 /* 1481 /*
1500 * \Sum shares * rq_weight 1482 * \Sum shares * rq_weight
@@ -1502,7 +1484,7 @@ update_group_shares_cpu(struct task_group *tg, int cpu,
1502 * \Sum rq_weight 1484 * \Sum rq_weight
1503 * 1485 *
1504 */ 1486 */
1505 shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); 1487 shares = (sd_shares * rq_weight) / sd_rq_weight;
1506 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); 1488 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
1507 1489
1508 if (abs(shares - tg->se[cpu]->load.weight) > 1490 if (abs(shares - tg->se[cpu]->load.weight) >
@@ -1511,11 +1493,7 @@ update_group_shares_cpu(struct task_group *tg, int cpu,
1511 unsigned long flags; 1493 unsigned long flags;
1512 1494
1513 spin_lock_irqsave(&rq->lock, flags); 1495 spin_lock_irqsave(&rq->lock, flags);
1514 /* 1496 tg->cfs_rq[cpu]->shares = shares;
1515 * record the actual number of shares, not the boosted amount.
1516 */
1517 tg->cfs_rq[cpu]->shares = boost ? 0 : shares;
1518 tg->cfs_rq[cpu]->rq_weight = rq_weight;
1519 1497
1520 __set_se_shares(tg->se[cpu], shares); 1498 __set_se_shares(tg->se[cpu], shares);
1521 spin_unlock_irqrestore(&rq->lock, flags); 1499 spin_unlock_irqrestore(&rq->lock, flags);
@@ -1529,13 +1507,23 @@ update_group_shares_cpu(struct task_group *tg, int cpu,
1529 */ 1507 */
1530static int tg_shares_up(struct task_group *tg, void *data) 1508static int tg_shares_up(struct task_group *tg, void *data)
1531{ 1509{
1532 unsigned long rq_weight = 0; 1510 unsigned long weight, rq_weight = 0;
1533 unsigned long shares = 0; 1511 unsigned long shares = 0;
1534 struct sched_domain *sd = data; 1512 struct sched_domain *sd = data;
1535 int i; 1513 int i;
1536 1514
1537 for_each_cpu_mask(i, sd->span) { 1515 for_each_cpu(i, sched_domain_span(sd)) {
1538 rq_weight += tg->cfs_rq[i]->load.weight; 1516 /*
1517 * If there are currently no tasks on the cpu pretend there
1518 * is one of average load so that when a new task gets to
1519 * run here it will not get delayed by group starvation.
1520 */
1521 weight = tg->cfs_rq[i]->load.weight;
1522 if (!weight)
1523 weight = NICE_0_LOAD;
1524
1525 tg->cfs_rq[i]->rq_weight = weight;
1526 rq_weight += weight;
1539 shares += tg->cfs_rq[i]->shares; 1527 shares += tg->cfs_rq[i]->shares;
1540 } 1528 }
1541 1529
@@ -1545,10 +1533,7 @@ static int tg_shares_up(struct task_group *tg, void *data)
1545 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) 1533 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1546 shares = tg->shares; 1534 shares = tg->shares;
1547 1535
1548 if (!rq_weight) 1536 for_each_cpu(i, sched_domain_span(sd))
1549 rq_weight = cpus_weight(sd->span) * NICE_0_LOAD;
1550
1551 for_each_cpu_mask(i, sd->span)
1552 update_group_shares_cpu(tg, i, shares, rq_weight); 1537 update_group_shares_cpu(tg, i, shares, rq_weight);
1553 1538
1554 return 0; 1539 return 0;
@@ -1612,6 +1597,39 @@ static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd)
1612 1597
1613#endif 1598#endif
1614 1599
1600/*
1601 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1602 */
1603static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1604 __releases(this_rq->lock)
1605 __acquires(busiest->lock)
1606 __acquires(this_rq->lock)
1607{
1608 int ret = 0;
1609
1610 if (unlikely(!irqs_disabled())) {
1611 /* printk() doesn't work good under rq->lock */
1612 spin_unlock(&this_rq->lock);
1613 BUG_ON(1);
1614 }
1615 if (unlikely(!spin_trylock(&busiest->lock))) {
1616 if (busiest < this_rq) {
1617 spin_unlock(&this_rq->lock);
1618 spin_lock(&busiest->lock);
1619 spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING);
1620 ret = 1;
1621 } else
1622 spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING);
1623 }
1624 return ret;
1625}
1626
1627static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1628 __releases(busiest->lock)
1629{
1630 spin_unlock(&busiest->lock);
1631 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1632}
1615#endif 1633#endif
1616 1634
1617#ifdef CONFIG_FAIR_GROUP_SCHED 1635#ifdef CONFIG_FAIR_GROUP_SCHED
@@ -2079,15 +2097,17 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
2079 int i; 2097 int i;
2080 2098
2081 /* Skip over this group if it has no CPUs allowed */ 2099 /* Skip over this group if it has no CPUs allowed */
2082 if (!cpus_intersects(group->cpumask, p->cpus_allowed)) 2100 if (!cpumask_intersects(sched_group_cpus(group),
2101 &p->cpus_allowed))
2083 continue; 2102 continue;
2084 2103
2085 local_group = cpu_isset(this_cpu, group->cpumask); 2104 local_group = cpumask_test_cpu(this_cpu,
2105 sched_group_cpus(group));
2086 2106
2087 /* Tally up the load of all CPUs in the group */ 2107 /* Tally up the load of all CPUs in the group */
2088 avg_load = 0; 2108 avg_load = 0;
2089 2109
2090 for_each_cpu_mask_nr(i, group->cpumask) { 2110 for_each_cpu(i, sched_group_cpus(group)) {
2091 /* Bias balancing toward cpus of our domain */ 2111 /* Bias balancing toward cpus of our domain */
2092 if (local_group) 2112 if (local_group)
2093 load = source_load(i, load_idx); 2113 load = source_load(i, load_idx);
@@ -2119,17 +2139,14 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
2119 * find_idlest_cpu - find the idlest cpu among the cpus in group. 2139 * find_idlest_cpu - find the idlest cpu among the cpus in group.
2120 */ 2140 */
2121static int 2141static int
2122find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, 2142find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
2123 cpumask_t *tmp)
2124{ 2143{
2125 unsigned long load, min_load = ULONG_MAX; 2144 unsigned long load, min_load = ULONG_MAX;
2126 int idlest = -1; 2145 int idlest = -1;
2127 int i; 2146 int i;
2128 2147
2129 /* Traverse only the allowed CPUs */ 2148 /* Traverse only the allowed CPUs */
2130 cpus_and(*tmp, group->cpumask, p->cpus_allowed); 2149 for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) {
2131
2132 for_each_cpu_mask_nr(i, *tmp) {
2133 load = weighted_cpuload(i); 2150 load = weighted_cpuload(i);
2134 2151
2135 if (load < min_load || (load == min_load && i == this_cpu)) { 2152 if (load < min_load || (load == min_load && i == this_cpu)) {
@@ -2171,7 +2188,6 @@ static int sched_balance_self(int cpu, int flag)
2171 update_shares(sd); 2188 update_shares(sd);
2172 2189
2173 while (sd) { 2190 while (sd) {
2174 cpumask_t span, tmpmask;
2175 struct sched_group *group; 2191 struct sched_group *group;
2176 int new_cpu, weight; 2192 int new_cpu, weight;
2177 2193
@@ -2180,14 +2196,13 @@ static int sched_balance_self(int cpu, int flag)
2180 continue; 2196 continue;
2181 } 2197 }
2182 2198
2183 span = sd->span;
2184 group = find_idlest_group(sd, t, cpu); 2199 group = find_idlest_group(sd, t, cpu);
2185 if (!group) { 2200 if (!group) {
2186 sd = sd->child; 2201 sd = sd->child;
2187 continue; 2202 continue;
2188 } 2203 }
2189 2204
2190 new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); 2205 new_cpu = find_idlest_cpu(group, t, cpu);
2191 if (new_cpu == -1 || new_cpu == cpu) { 2206 if (new_cpu == -1 || new_cpu == cpu) {
2192 /* Now try balancing at a lower domain level of cpu */ 2207 /* Now try balancing at a lower domain level of cpu */
2193 sd = sd->child; 2208 sd = sd->child;
@@ -2196,10 +2211,10 @@ static int sched_balance_self(int cpu, int flag)
2196 2211
2197 /* Now try balancing at a lower domain level of new_cpu */ 2212 /* Now try balancing at a lower domain level of new_cpu */
2198 cpu = new_cpu; 2213 cpu = new_cpu;
2214 weight = cpumask_weight(sched_domain_span(sd));
2199 sd = NULL; 2215 sd = NULL;
2200 weight = cpus_weight(span);
2201 for_each_domain(cpu, tmp) { 2216 for_each_domain(cpu, tmp) {
2202 if (weight <= cpus_weight(tmp->span)) 2217 if (weight <= cpumask_weight(sched_domain_span(tmp)))
2203 break; 2218 break;
2204 if (tmp->flags & flag) 2219 if (tmp->flags & flag)
2205 sd = tmp; 2220 sd = tmp;
@@ -2244,7 +2259,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2244 cpu = task_cpu(p); 2259 cpu = task_cpu(p);
2245 2260
2246 for_each_domain(this_cpu, sd) { 2261 for_each_domain(this_cpu, sd) {
2247 if (cpu_isset(cpu, sd->span)) { 2262 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2248 update_shares(sd); 2263 update_shares(sd);
2249 break; 2264 break;
2250 } 2265 }
@@ -2292,7 +2307,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2292 else { 2307 else {
2293 struct sched_domain *sd; 2308 struct sched_domain *sd;
2294 for_each_domain(this_cpu, sd) { 2309 for_each_domain(this_cpu, sd) {
2295 if (cpu_isset(cpu, sd->span)) { 2310 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2296 schedstat_inc(sd, ttwu_wake_remote); 2311 schedstat_inc(sd, ttwu_wake_remote);
2297 break; 2312 break;
2298 } 2313 }
@@ -2812,40 +2827,6 @@ static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2812} 2827}
2813 2828
2814/* 2829/*
2815 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
2816 */
2817static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
2818 __releases(this_rq->lock)
2819 __acquires(busiest->lock)
2820 __acquires(this_rq->lock)
2821{
2822 int ret = 0;
2823
2824 if (unlikely(!irqs_disabled())) {
2825 /* printk() doesn't work good under rq->lock */
2826 spin_unlock(&this_rq->lock);
2827 BUG_ON(1);
2828 }
2829 if (unlikely(!spin_trylock(&busiest->lock))) {
2830 if (busiest < this_rq) {
2831 spin_unlock(&this_rq->lock);
2832 spin_lock(&busiest->lock);
2833 spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING);
2834 ret = 1;
2835 } else
2836 spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING);
2837 }
2838 return ret;
2839}
2840
2841static void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
2842 __releases(busiest->lock)
2843{
2844 spin_unlock(&busiest->lock);
2845 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
2846}
2847
2848/*
2849 * If dest_cpu is allowed for this process, migrate the task to it. 2830 * If dest_cpu is allowed for this process, migrate the task to it.
2850 * This is accomplished by forcing the cpu_allowed mask to only 2831 * This is accomplished by forcing the cpu_allowed mask to only
2851 * allow dest_cpu, which will force the cpu onto dest_cpu. Then 2832 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
@@ -2858,7 +2839,7 @@ static void sched_migrate_task(struct task_struct *p, int dest_cpu)
2858 struct rq *rq; 2839 struct rq *rq;
2859 2840
2860 rq = task_rq_lock(p, &flags); 2841 rq = task_rq_lock(p, &flags);
2861 if (!cpu_isset(dest_cpu, p->cpus_allowed) 2842 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)
2862 || unlikely(!cpu_active(dest_cpu))) 2843 || unlikely(!cpu_active(dest_cpu)))
2863 goto out; 2844 goto out;
2864 2845
@@ -2924,7 +2905,7 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
2924 * 2) cannot be migrated to this CPU due to cpus_allowed, or 2905 * 2) cannot be migrated to this CPU due to cpus_allowed, or
2925 * 3) are cache-hot on their current CPU. 2906 * 3) are cache-hot on their current CPU.
2926 */ 2907 */
2927 if (!cpu_isset(this_cpu, p->cpus_allowed)) { 2908 if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) {
2928 schedstat_inc(p, se.nr_failed_migrations_affine); 2909 schedstat_inc(p, se.nr_failed_migrations_affine);
2929 return 0; 2910 return 0;
2930 } 2911 }
@@ -3099,7 +3080,7 @@ static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
3099static struct sched_group * 3080static struct sched_group *
3100find_busiest_group(struct sched_domain *sd, int this_cpu, 3081find_busiest_group(struct sched_domain *sd, int this_cpu,
3101 unsigned long *imbalance, enum cpu_idle_type idle, 3082 unsigned long *imbalance, enum cpu_idle_type idle,
3102 int *sd_idle, const cpumask_t *cpus, int *balance) 3083 int *sd_idle, const struct cpumask *cpus, int *balance)
3103{ 3084{
3104 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; 3085 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
3105 unsigned long max_load, avg_load, total_load, this_load, total_pwr; 3086 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
@@ -3135,10 +3116,11 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3135 unsigned long sum_avg_load_per_task; 3116 unsigned long sum_avg_load_per_task;
3136 unsigned long avg_load_per_task; 3117 unsigned long avg_load_per_task;
3137 3118
3138 local_group = cpu_isset(this_cpu, group->cpumask); 3119 local_group = cpumask_test_cpu(this_cpu,
3120 sched_group_cpus(group));
3139 3121
3140 if (local_group) 3122 if (local_group)
3141 balance_cpu = first_cpu(group->cpumask); 3123 balance_cpu = cpumask_first(sched_group_cpus(group));
3142 3124
3143 /* Tally up the load of all CPUs in the group */ 3125 /* Tally up the load of all CPUs in the group */
3144 sum_weighted_load = sum_nr_running = avg_load = 0; 3126 sum_weighted_load = sum_nr_running = avg_load = 0;
@@ -3147,13 +3129,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3147 max_cpu_load = 0; 3129 max_cpu_load = 0;
3148 min_cpu_load = ~0UL; 3130 min_cpu_load = ~0UL;
3149 3131
3150 for_each_cpu_mask_nr(i, group->cpumask) { 3132 for_each_cpu_and(i, sched_group_cpus(group), cpus) {
3151 struct rq *rq; 3133 struct rq *rq = cpu_rq(i);
3152
3153 if (!cpu_isset(i, *cpus))
3154 continue;
3155
3156 rq = cpu_rq(i);
3157 3134
3158 if (*sd_idle && rq->nr_running) 3135 if (*sd_idle && rq->nr_running)
3159 *sd_idle = 0; 3136 *sd_idle = 0;
@@ -3264,8 +3241,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3264 */ 3241 */
3265 if ((sum_nr_running < min_nr_running) || 3242 if ((sum_nr_running < min_nr_running) ||
3266 (sum_nr_running == min_nr_running && 3243 (sum_nr_running == min_nr_running &&
3267 first_cpu(group->cpumask) < 3244 cpumask_first(sched_group_cpus(group)) <
3268 first_cpu(group_min->cpumask))) { 3245 cpumask_first(sched_group_cpus(group_min)))) {
3269 group_min = group; 3246 group_min = group;
3270 min_nr_running = sum_nr_running; 3247 min_nr_running = sum_nr_running;
3271 min_load_per_task = sum_weighted_load / 3248 min_load_per_task = sum_weighted_load /
@@ -3280,8 +3257,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3280 if (sum_nr_running <= group_capacity - 1) { 3257 if (sum_nr_running <= group_capacity - 1) {
3281 if (sum_nr_running > leader_nr_running || 3258 if (sum_nr_running > leader_nr_running ||
3282 (sum_nr_running == leader_nr_running && 3259 (sum_nr_running == leader_nr_running &&
3283 first_cpu(group->cpumask) > 3260 cpumask_first(sched_group_cpus(group)) >
3284 first_cpu(group_leader->cpumask))) { 3261 cpumask_first(sched_group_cpus(group_leader)))) {
3285 group_leader = group; 3262 group_leader = group;
3286 leader_nr_running = sum_nr_running; 3263 leader_nr_running = sum_nr_running;
3287 } 3264 }
@@ -3420,16 +3397,16 @@ ret:
3420 */ 3397 */
3421static struct rq * 3398static struct rq *
3422find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, 3399find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
3423 unsigned long imbalance, const cpumask_t *cpus) 3400 unsigned long imbalance, const struct cpumask *cpus)
3424{ 3401{
3425 struct rq *busiest = NULL, *rq; 3402 struct rq *busiest = NULL, *rq;
3426 unsigned long max_load = 0; 3403 unsigned long max_load = 0;
3427 int i; 3404 int i;
3428 3405
3429 for_each_cpu_mask_nr(i, group->cpumask) { 3406 for_each_cpu(i, sched_group_cpus(group)) {
3430 unsigned long wl; 3407 unsigned long wl;
3431 3408
3432 if (!cpu_isset(i, *cpus)) 3409 if (!cpumask_test_cpu(i, cpus))
3433 continue; 3410 continue;
3434 3411
3435 rq = cpu_rq(i); 3412 rq = cpu_rq(i);
@@ -3459,7 +3436,7 @@ find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
3459 */ 3436 */
3460static int load_balance(int this_cpu, struct rq *this_rq, 3437static int load_balance(int this_cpu, struct rq *this_rq,
3461 struct sched_domain *sd, enum cpu_idle_type idle, 3438 struct sched_domain *sd, enum cpu_idle_type idle,
3462 int *balance, cpumask_t *cpus) 3439 int *balance, struct cpumask *cpus)
3463{ 3440{
3464 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; 3441 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
3465 struct sched_group *group; 3442 struct sched_group *group;
@@ -3467,7 +3444,7 @@ static int load_balance(int this_cpu, struct rq *this_rq,
3467 struct rq *busiest; 3444 struct rq *busiest;
3468 unsigned long flags; 3445 unsigned long flags;
3469 3446
3470 cpus_setall(*cpus); 3447 cpumask_setall(cpus);
3471 3448
3472 /* 3449 /*
3473 * When power savings policy is enabled for the parent domain, idle 3450 * When power savings policy is enabled for the parent domain, idle
@@ -3527,8 +3504,8 @@ redo:
3527 3504
3528 /* All tasks on this runqueue were pinned by CPU affinity */ 3505 /* All tasks on this runqueue were pinned by CPU affinity */
3529 if (unlikely(all_pinned)) { 3506 if (unlikely(all_pinned)) {
3530 cpu_clear(cpu_of(busiest), *cpus); 3507 cpumask_clear_cpu(cpu_of(busiest), cpus);
3531 if (!cpus_empty(*cpus)) 3508 if (!cpumask_empty(cpus))
3532 goto redo; 3509 goto redo;
3533 goto out_balanced; 3510 goto out_balanced;
3534 } 3511 }
@@ -3545,7 +3522,8 @@ redo:
3545 /* don't kick the migration_thread, if the curr 3522 /* don't kick the migration_thread, if the curr
3546 * task on busiest cpu can't be moved to this_cpu 3523 * task on busiest cpu can't be moved to this_cpu
3547 */ 3524 */
3548 if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { 3525 if (!cpumask_test_cpu(this_cpu,
3526 &busiest->curr->cpus_allowed)) {
3549 spin_unlock_irqrestore(&busiest->lock, flags); 3527 spin_unlock_irqrestore(&busiest->lock, flags);
3550 all_pinned = 1; 3528 all_pinned = 1;
3551 goto out_one_pinned; 3529 goto out_one_pinned;
@@ -3620,7 +3598,7 @@ out:
3620 */ 3598 */
3621static int 3599static int
3622load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, 3600load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd,
3623 cpumask_t *cpus) 3601 struct cpumask *cpus)
3624{ 3602{
3625 struct sched_group *group; 3603 struct sched_group *group;
3626 struct rq *busiest = NULL; 3604 struct rq *busiest = NULL;
@@ -3629,7 +3607,7 @@ load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd,
3629 int sd_idle = 0; 3607 int sd_idle = 0;
3630 int all_pinned = 0; 3608 int all_pinned = 0;
3631 3609
3632 cpus_setall(*cpus); 3610 cpumask_setall(cpus);
3633 3611
3634 /* 3612 /*
3635 * When power savings policy is enabled for the parent domain, idle 3613 * When power savings policy is enabled for the parent domain, idle
@@ -3673,8 +3651,8 @@ redo:
3673 double_unlock_balance(this_rq, busiest); 3651 double_unlock_balance(this_rq, busiest);
3674 3652
3675 if (unlikely(all_pinned)) { 3653 if (unlikely(all_pinned)) {
3676 cpu_clear(cpu_of(busiest), *cpus); 3654 cpumask_clear_cpu(cpu_of(busiest), cpus);
3677 if (!cpus_empty(*cpus)) 3655 if (!cpumask_empty(cpus))
3678 goto redo; 3656 goto redo;
3679 } 3657 }
3680 } 3658 }
@@ -3707,9 +3685,12 @@ out_balanced:
3707static void idle_balance(int this_cpu, struct rq *this_rq) 3685static void idle_balance(int this_cpu, struct rq *this_rq)
3708{ 3686{
3709 struct sched_domain *sd; 3687 struct sched_domain *sd;
3710 int pulled_task = -1; 3688 int pulled_task = 0;
3711 unsigned long next_balance = jiffies + HZ; 3689 unsigned long next_balance = jiffies + HZ;
3712 cpumask_t tmpmask; 3690 cpumask_var_t tmpmask;
3691
3692 if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC))
3693 return;
3713 3694
3714 for_each_domain(this_cpu, sd) { 3695 for_each_domain(this_cpu, sd) {
3715 unsigned long interval; 3696 unsigned long interval;
@@ -3720,7 +3701,7 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3720 if (sd->flags & SD_BALANCE_NEWIDLE) 3701 if (sd->flags & SD_BALANCE_NEWIDLE)
3721 /* If we've pulled tasks over stop searching: */ 3702 /* If we've pulled tasks over stop searching: */
3722 pulled_task = load_balance_newidle(this_cpu, this_rq, 3703 pulled_task = load_balance_newidle(this_cpu, this_rq,
3723 sd, &tmpmask); 3704 sd, tmpmask);
3724 3705
3725 interval = msecs_to_jiffies(sd->balance_interval); 3706 interval = msecs_to_jiffies(sd->balance_interval);
3726 if (time_after(next_balance, sd->last_balance + interval)) 3707 if (time_after(next_balance, sd->last_balance + interval))
@@ -3735,6 +3716,7 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3735 */ 3716 */
3736 this_rq->next_balance = next_balance; 3717 this_rq->next_balance = next_balance;
3737 } 3718 }
3719 free_cpumask_var(tmpmask);
3738} 3720}
3739 3721
3740/* 3722/*
@@ -3772,7 +3754,7 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
3772 /* Search for an sd spanning us and the target CPU. */ 3754 /* Search for an sd spanning us and the target CPU. */
3773 for_each_domain(target_cpu, sd) { 3755 for_each_domain(target_cpu, sd) {
3774 if ((sd->flags & SD_LOAD_BALANCE) && 3756 if ((sd->flags & SD_LOAD_BALANCE) &&
3775 cpu_isset(busiest_cpu, sd->span)) 3757 cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
3776 break; 3758 break;
3777 } 3759 }
3778 3760
@@ -3791,10 +3773,9 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
3791#ifdef CONFIG_NO_HZ 3773#ifdef CONFIG_NO_HZ
3792static struct { 3774static struct {
3793 atomic_t load_balancer; 3775 atomic_t load_balancer;
3794 cpumask_t cpu_mask; 3776 cpumask_var_t cpu_mask;
3795} nohz ____cacheline_aligned = { 3777} nohz ____cacheline_aligned = {
3796 .load_balancer = ATOMIC_INIT(-1), 3778 .load_balancer = ATOMIC_INIT(-1),
3797 .cpu_mask = CPU_MASK_NONE,
3798}; 3779};
3799 3780
3800/* 3781/*
@@ -3822,7 +3803,7 @@ int select_nohz_load_balancer(int stop_tick)
3822 int cpu = smp_processor_id(); 3803 int cpu = smp_processor_id();
3823 3804
3824 if (stop_tick) { 3805 if (stop_tick) {
3825 cpu_set(cpu, nohz.cpu_mask); 3806 cpumask_set_cpu(cpu, nohz.cpu_mask);
3826 cpu_rq(cpu)->in_nohz_recently = 1; 3807 cpu_rq(cpu)->in_nohz_recently = 1;
3827 3808
3828 /* 3809 /*
@@ -3836,7 +3817,7 @@ int select_nohz_load_balancer(int stop_tick)
3836 } 3817 }
3837 3818
3838 /* time for ilb owner also to sleep */ 3819 /* time for ilb owner also to sleep */
3839 if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { 3820 if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) {
3840 if (atomic_read(&nohz.load_balancer) == cpu) 3821 if (atomic_read(&nohz.load_balancer) == cpu)
3841 atomic_set(&nohz.load_balancer, -1); 3822 atomic_set(&nohz.load_balancer, -1);
3842 return 0; 3823 return 0;
@@ -3849,10 +3830,10 @@ int select_nohz_load_balancer(int stop_tick)
3849 } else if (atomic_read(&nohz.load_balancer) == cpu) 3830 } else if (atomic_read(&nohz.load_balancer) == cpu)
3850 return 1; 3831 return 1;
3851 } else { 3832 } else {
3852 if (!cpu_isset(cpu, nohz.cpu_mask)) 3833 if (!cpumask_test_cpu(cpu, nohz.cpu_mask))
3853 return 0; 3834 return 0;
3854 3835
3855 cpu_clear(cpu, nohz.cpu_mask); 3836 cpumask_clear_cpu(cpu, nohz.cpu_mask);
3856 3837
3857 if (atomic_read(&nohz.load_balancer) == cpu) 3838 if (atomic_read(&nohz.load_balancer) == cpu)
3858 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) 3839 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
@@ -3880,7 +3861,11 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
3880 unsigned long next_balance = jiffies + 60*HZ; 3861 unsigned long next_balance = jiffies + 60*HZ;
3881 int update_next_balance = 0; 3862 int update_next_balance = 0;
3882 int need_serialize; 3863 int need_serialize;
3883 cpumask_t tmp; 3864 cpumask_var_t tmp;
3865
3866 /* Fails alloc? Rebalancing probably not a priority right now. */
3867 if (!alloc_cpumask_var(&tmp, GFP_ATOMIC))
3868 return;
3884 3869
3885 for_each_domain(cpu, sd) { 3870 for_each_domain(cpu, sd) {
3886 if (!(sd->flags & SD_LOAD_BALANCE)) 3871 if (!(sd->flags & SD_LOAD_BALANCE))
@@ -3905,7 +3890,7 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
3905 } 3890 }
3906 3891
3907 if (time_after_eq(jiffies, sd->last_balance + interval)) { 3892 if (time_after_eq(jiffies, sd->last_balance + interval)) {
3908 if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { 3893 if (load_balance(cpu, rq, sd, idle, &balance, tmp)) {
3909 /* 3894 /*
3910 * We've pulled tasks over so either we're no 3895 * We've pulled tasks over so either we're no
3911 * longer idle, or one of our SMT siblings is 3896 * longer idle, or one of our SMT siblings is
@@ -3939,6 +3924,8 @@ out:
3939 */ 3924 */
3940 if (likely(update_next_balance)) 3925 if (likely(update_next_balance))
3941 rq->next_balance = next_balance; 3926 rq->next_balance = next_balance;
3927
3928 free_cpumask_var(tmp);
3942} 3929}
3943 3930
3944/* 3931/*
@@ -3963,12 +3950,13 @@ static void run_rebalance_domains(struct softirq_action *h)
3963 */ 3950 */
3964 if (this_rq->idle_at_tick && 3951 if (this_rq->idle_at_tick &&
3965 atomic_read(&nohz.load_balancer) == this_cpu) { 3952 atomic_read(&nohz.load_balancer) == this_cpu) {
3966 cpumask_t cpus = nohz.cpu_mask;
3967 struct rq *rq; 3953 struct rq *rq;
3968 int balance_cpu; 3954 int balance_cpu;
3969 3955
3970 cpu_clear(this_cpu, cpus); 3956 for_each_cpu(balance_cpu, nohz.cpu_mask) {
3971 for_each_cpu_mask_nr(balance_cpu, cpus) { 3957 if (balance_cpu == this_cpu)
3958 continue;
3959
3972 /* 3960 /*
3973 * If this cpu gets work to do, stop the load balancing 3961 * If this cpu gets work to do, stop the load balancing
3974 * work being done for other cpus. Next load 3962 * work being done for other cpus. Next load
@@ -4006,7 +3994,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4006 rq->in_nohz_recently = 0; 3994 rq->in_nohz_recently = 0;
4007 3995
4008 if (atomic_read(&nohz.load_balancer) == cpu) { 3996 if (atomic_read(&nohz.load_balancer) == cpu) {
4009 cpu_clear(cpu, nohz.cpu_mask); 3997 cpumask_clear_cpu(cpu, nohz.cpu_mask);
4010 atomic_set(&nohz.load_balancer, -1); 3998 atomic_set(&nohz.load_balancer, -1);
4011 } 3999 }
4012 4000
@@ -4019,7 +4007,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4019 * TBD: Traverse the sched domains and nominate 4007 * TBD: Traverse the sched domains and nominate
4020 * the nearest cpu in the nohz.cpu_mask. 4008 * the nearest cpu in the nohz.cpu_mask.
4021 */ 4009 */
4022 int ilb = first_cpu(nohz.cpu_mask); 4010 int ilb = cpumask_first(nohz.cpu_mask);
4023 4011
4024 if (ilb < nr_cpu_ids) 4012 if (ilb < nr_cpu_ids)
4025 resched_cpu(ilb); 4013 resched_cpu(ilb);
@@ -4031,7 +4019,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4031 * cpus with ticks stopped, is it time for that to stop? 4019 * cpus with ticks stopped, is it time for that to stop?
4032 */ 4020 */
4033 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && 4021 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu &&
4034 cpus_weight(nohz.cpu_mask) == num_online_cpus()) { 4022 cpumask_weight(nohz.cpu_mask) == num_online_cpus()) {
4035 resched_cpu(cpu); 4023 resched_cpu(cpu);
4036 return; 4024 return;
4037 } 4025 }
@@ -4041,7 +4029,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4041 * someone else, then no need raise the SCHED_SOFTIRQ 4029 * someone else, then no need raise the SCHED_SOFTIRQ
4042 */ 4030 */
4043 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && 4031 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu &&
4044 cpu_isset(cpu, nohz.cpu_mask)) 4032 cpumask_test_cpu(cpu, nohz.cpu_mask))
4045 return; 4033 return;
4046#endif 4034#endif
4047 if (time_after_eq(jiffies, rq->next_balance)) 4035 if (time_after_eq(jiffies, rq->next_balance))
@@ -4203,7 +4191,6 @@ void account_steal_time(struct task_struct *p, cputime_t steal)
4203 4191
4204 if (p == rq->idle) { 4192 if (p == rq->idle) {
4205 p->stime = cputime_add(p->stime, steal); 4193 p->stime = cputime_add(p->stime, steal);
4206 account_group_system_time(p, steal);
4207 if (atomic_read(&rq->nr_iowait) > 0) 4194 if (atomic_read(&rq->nr_iowait) > 0)
4208 cpustat->iowait = cputime64_add(cpustat->iowait, tmp); 4195 cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
4209 else 4196 else
@@ -4339,7 +4326,7 @@ void __kprobes sub_preempt_count(int val)
4339 /* 4326 /*
4340 * Underflow? 4327 * Underflow?
4341 */ 4328 */
4342 if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) 4329 if (DEBUG_LOCKS_WARN_ON(val > preempt_count() - (!!kernel_locked())))
4343 return; 4330 return;
4344 /* 4331 /*
4345 * Is the spinlock portion underflowing? 4332 * Is the spinlock portion underflowing?
@@ -5400,10 +5387,9 @@ out_unlock:
5400 return retval; 5387 return retval;
5401} 5388}
5402 5389
5403long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) 5390long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
5404{ 5391{
5405 cpumask_t cpus_allowed; 5392 cpumask_var_t cpus_allowed, new_mask;
5406 cpumask_t new_mask = *in_mask;
5407 struct task_struct *p; 5393 struct task_struct *p;
5408 int retval; 5394 int retval;
5409 5395
@@ -5425,6 +5411,14 @@ long sched_setaffinity(pid_t pid, const cpumask_t *in_mask)
5425 get_task_struct(p); 5411 get_task_struct(p);
5426 read_unlock(&tasklist_lock); 5412 read_unlock(&tasklist_lock);
5427 5413
5414 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5415 retval = -ENOMEM;
5416 goto out_put_task;
5417 }
5418 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5419 retval = -ENOMEM;
5420 goto out_free_cpus_allowed;
5421 }
5428 retval = -EPERM; 5422 retval = -EPERM;
5429 if ((current->euid != p->euid) && (current->euid != p->uid) && 5423 if ((current->euid != p->euid) && (current->euid != p->uid) &&
5430 !capable(CAP_SYS_NICE)) 5424 !capable(CAP_SYS_NICE))
@@ -5434,37 +5428,41 @@ long sched_setaffinity(pid_t pid, const cpumask_t *in_mask)
5434 if (retval) 5428 if (retval)
5435 goto out_unlock; 5429 goto out_unlock;
5436 5430
5437 cpuset_cpus_allowed(p, &cpus_allowed); 5431 cpuset_cpus_allowed(p, cpus_allowed);
5438 cpus_and(new_mask, new_mask, cpus_allowed); 5432 cpumask_and(new_mask, in_mask, cpus_allowed);
5439 again: 5433 again:
5440 retval = set_cpus_allowed_ptr(p, &new_mask); 5434 retval = set_cpus_allowed_ptr(p, new_mask);
5441 5435
5442 if (!retval) { 5436 if (!retval) {
5443 cpuset_cpus_allowed(p, &cpus_allowed); 5437 cpuset_cpus_allowed(p, cpus_allowed);
5444 if (!cpus_subset(new_mask, cpus_allowed)) { 5438 if (!cpumask_subset(new_mask, cpus_allowed)) {
5445 /* 5439 /*
5446 * We must have raced with a concurrent cpuset 5440 * We must have raced with a concurrent cpuset
5447 * update. Just reset the cpus_allowed to the 5441 * update. Just reset the cpus_allowed to the
5448 * cpuset's cpus_allowed 5442 * cpuset's cpus_allowed
5449 */ 5443 */
5450 new_mask = cpus_allowed; 5444 cpumask_copy(new_mask, cpus_allowed);
5451 goto again; 5445 goto again;
5452 } 5446 }
5453 } 5447 }
5454out_unlock: 5448out_unlock:
5449 free_cpumask_var(new_mask);
5450out_free_cpus_allowed:
5451 free_cpumask_var(cpus_allowed);
5452out_put_task:
5455 put_task_struct(p); 5453 put_task_struct(p);
5456 put_online_cpus(); 5454 put_online_cpus();
5457 return retval; 5455 return retval;
5458} 5456}
5459 5457
5460static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, 5458static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
5461 cpumask_t *new_mask) 5459 struct cpumask *new_mask)
5462{ 5460{
5463 if (len < sizeof(cpumask_t)) { 5461 if (len < cpumask_size())
5464 memset(new_mask, 0, sizeof(cpumask_t)); 5462 cpumask_clear(new_mask);
5465 } else if (len > sizeof(cpumask_t)) { 5463 else if (len > cpumask_size())
5466 len = sizeof(cpumask_t); 5464 len = cpumask_size();
5467 } 5465
5468 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; 5466 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5469} 5467}
5470 5468
@@ -5477,17 +5475,20 @@ static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
5477asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, 5475asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
5478 unsigned long __user *user_mask_ptr) 5476 unsigned long __user *user_mask_ptr)
5479{ 5477{
5480 cpumask_t new_mask; 5478 cpumask_var_t new_mask;
5481 int retval; 5479 int retval;
5482 5480
5483 retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); 5481 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5484 if (retval) 5482 return -ENOMEM;
5485 return retval;
5486 5483
5487 return sched_setaffinity(pid, &new_mask); 5484 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5485 if (retval == 0)
5486 retval = sched_setaffinity(pid, new_mask);
5487 free_cpumask_var(new_mask);
5488 return retval;
5488} 5489}
5489 5490
5490long sched_getaffinity(pid_t pid, cpumask_t *mask) 5491long sched_getaffinity(pid_t pid, struct cpumask *mask)
5491{ 5492{
5492 struct task_struct *p; 5493 struct task_struct *p;
5493 int retval; 5494 int retval;
@@ -5504,7 +5505,7 @@ long sched_getaffinity(pid_t pid, cpumask_t *mask)
5504 if (retval) 5505 if (retval)
5505 goto out_unlock; 5506 goto out_unlock;
5506 5507
5507 cpus_and(*mask, p->cpus_allowed, cpu_online_map); 5508 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
5508 5509
5509out_unlock: 5510out_unlock:
5510 read_unlock(&tasklist_lock); 5511 read_unlock(&tasklist_lock);
@@ -5523,19 +5524,24 @@ asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
5523 unsigned long __user *user_mask_ptr) 5524 unsigned long __user *user_mask_ptr)
5524{ 5525{
5525 int ret; 5526 int ret;
5526 cpumask_t mask; 5527 cpumask_var_t mask;
5527 5528
5528 if (len < sizeof(cpumask_t)) 5529 if (len < cpumask_size())
5529 return -EINVAL; 5530 return -EINVAL;
5530 5531
5531 ret = sched_getaffinity(pid, &mask); 5532 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5532 if (ret < 0) 5533 return -ENOMEM;
5533 return ret;
5534 5534
5535 if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) 5535 ret = sched_getaffinity(pid, mask);
5536 return -EFAULT; 5536 if (ret == 0) {
5537 if (copy_to_user(user_mask_ptr, mask, cpumask_size()))
5538 ret = -EFAULT;
5539 else
5540 ret = cpumask_size();
5541 }
5542 free_cpumask_var(mask);
5537 5543
5538 return sizeof(cpumask_t); 5544 return ret;
5539} 5545}
5540 5546
5541/** 5547/**
@@ -5877,7 +5883,7 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5877 idle->se.exec_start = sched_clock(); 5883 idle->se.exec_start = sched_clock();
5878 5884
5879 idle->prio = idle->normal_prio = MAX_PRIO; 5885 idle->prio = idle->normal_prio = MAX_PRIO;
5880 idle->cpus_allowed = cpumask_of_cpu(cpu); 5886 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
5881 __set_task_cpu(idle, cpu); 5887 __set_task_cpu(idle, cpu);
5882 5888
5883 rq->curr = rq->idle = idle; 5889 rq->curr = rq->idle = idle;
@@ -5896,6 +5902,7 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5896 * The idle tasks have their own, simple scheduling class: 5902 * The idle tasks have their own, simple scheduling class:
5897 */ 5903 */
5898 idle->sched_class = &idle_sched_class; 5904 idle->sched_class = &idle_sched_class;
5905 ftrace_graph_init_task(idle);
5899} 5906}
5900 5907
5901/* 5908/*
@@ -5903,9 +5910,9 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5903 * indicates which cpus entered this state. This is used 5910 * indicates which cpus entered this state. This is used
5904 * in the rcu update to wait only for active cpus. For system 5911 * in the rcu update to wait only for active cpus. For system
5905 * which do not switch off the HZ timer nohz_cpu_mask should 5912 * which do not switch off the HZ timer nohz_cpu_mask should
5906 * always be CPU_MASK_NONE. 5913 * always be CPU_BITS_NONE.
5907 */ 5914 */
5908cpumask_t nohz_cpu_mask = CPU_MASK_NONE; 5915cpumask_var_t nohz_cpu_mask;
5909 5916
5910/* 5917/*
5911 * Increase the granularity value when there are more CPUs, 5918 * Increase the granularity value when there are more CPUs,
@@ -5960,7 +5967,7 @@ static inline void sched_init_granularity(void)
5960 * task must not exit() & deallocate itself prematurely. The 5967 * task must not exit() & deallocate itself prematurely. The
5961 * call is not atomic; no spinlocks may be held. 5968 * call is not atomic; no spinlocks may be held.
5962 */ 5969 */
5963int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) 5970int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
5964{ 5971{
5965 struct migration_req req; 5972 struct migration_req req;
5966 unsigned long flags; 5973 unsigned long flags;
@@ -5968,13 +5975,13 @@ int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask)
5968 int ret = 0; 5975 int ret = 0;
5969 5976
5970 rq = task_rq_lock(p, &flags); 5977 rq = task_rq_lock(p, &flags);
5971 if (!cpus_intersects(*new_mask, cpu_online_map)) { 5978 if (!cpumask_intersects(new_mask, cpu_online_mask)) {
5972 ret = -EINVAL; 5979 ret = -EINVAL;
5973 goto out; 5980 goto out;
5974 } 5981 }
5975 5982
5976 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && 5983 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
5977 !cpus_equal(p->cpus_allowed, *new_mask))) { 5984 !cpumask_equal(&p->cpus_allowed, new_mask))) {
5978 ret = -EINVAL; 5985 ret = -EINVAL;
5979 goto out; 5986 goto out;
5980 } 5987 }
@@ -5982,15 +5989,15 @@ int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask)
5982 if (p->sched_class->set_cpus_allowed) 5989 if (p->sched_class->set_cpus_allowed)
5983 p->sched_class->set_cpus_allowed(p, new_mask); 5990 p->sched_class->set_cpus_allowed(p, new_mask);
5984 else { 5991 else {
5985 p->cpus_allowed = *new_mask; 5992 cpumask_copy(&p->cpus_allowed, new_mask);
5986 p->rt.nr_cpus_allowed = cpus_weight(*new_mask); 5993 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
5987 } 5994 }
5988 5995
5989 /* Can the task run on the task's current CPU? If so, we're done */ 5996 /* Can the task run on the task's current CPU? If so, we're done */
5990 if (cpu_isset(task_cpu(p), *new_mask)) 5997 if (cpumask_test_cpu(task_cpu(p), new_mask))
5991 goto out; 5998 goto out;
5992 5999
5993 if (migrate_task(p, any_online_cpu(*new_mask), &req)) { 6000 if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) {
5994 /* Need help from migration thread: drop lock and wait. */ 6001 /* Need help from migration thread: drop lock and wait. */
5995 task_rq_unlock(rq, &flags); 6002 task_rq_unlock(rq, &flags);
5996 wake_up_process(rq->migration_thread); 6003 wake_up_process(rq->migration_thread);
@@ -6032,7 +6039,7 @@ static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
6032 if (task_cpu(p) != src_cpu) 6039 if (task_cpu(p) != src_cpu)
6033 goto done; 6040 goto done;
6034 /* Affinity changed (again). */ 6041 /* Affinity changed (again). */
6035 if (!cpu_isset(dest_cpu, p->cpus_allowed)) 6042 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
6036 goto fail; 6043 goto fail;
6037 6044
6038 on_rq = p->se.on_rq; 6045 on_rq = p->se.on_rq;
@@ -6126,54 +6133,46 @@ static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
6126 6133
6127/* 6134/*
6128 * Figure out where task on dead CPU should go, use force if necessary. 6135 * Figure out where task on dead CPU should go, use force if necessary.
6129 * NOTE: interrupts should be disabled by the caller
6130 */ 6136 */
6131static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) 6137static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
6132{ 6138{
6133 unsigned long flags;
6134 cpumask_t mask;
6135 struct rq *rq;
6136 int dest_cpu; 6139 int dest_cpu;
6140 /* FIXME: Use cpumask_of_node here. */
6141 cpumask_t _nodemask = node_to_cpumask(cpu_to_node(dead_cpu));
6142 const struct cpumask *nodemask = &_nodemask;
6143
6144again:
6145 /* Look for allowed, online CPU in same node. */
6146 for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask)
6147 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
6148 goto move;
6149
6150 /* Any allowed, online CPU? */
6151 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask);
6152 if (dest_cpu < nr_cpu_ids)
6153 goto move;
6154
6155 /* No more Mr. Nice Guy. */
6156 if (dest_cpu >= nr_cpu_ids) {
6157 cpuset_cpus_allowed_locked(p, &p->cpus_allowed);
6158 dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed);
6137 6159
6138 do { 6160 /*
6139 /* On same node? */ 6161 * Don't tell them about moving exiting tasks or
6140 mask = node_to_cpumask(cpu_to_node(dead_cpu)); 6162 * kernel threads (both mm NULL), since they never
6141 cpus_and(mask, mask, p->cpus_allowed); 6163 * leave kernel.
6142 dest_cpu = any_online_cpu(mask); 6164 */
6143 6165 if (p->mm && printk_ratelimit()) {
6144 /* On any allowed CPU? */ 6166 printk(KERN_INFO "process %d (%s) no "
6145 if (dest_cpu >= nr_cpu_ids) 6167 "longer affine to cpu%d\n",
6146 dest_cpu = any_online_cpu(p->cpus_allowed); 6168 task_pid_nr(p), p->comm, dead_cpu);
6147
6148 /* No more Mr. Nice Guy. */
6149 if (dest_cpu >= nr_cpu_ids) {
6150 cpumask_t cpus_allowed;
6151
6152 cpuset_cpus_allowed_locked(p, &cpus_allowed);
6153 /*
6154 * Try to stay on the same cpuset, where the
6155 * current cpuset may be a subset of all cpus.
6156 * The cpuset_cpus_allowed_locked() variant of
6157 * cpuset_cpus_allowed() will not block. It must be
6158 * called within calls to cpuset_lock/cpuset_unlock.
6159 */
6160 rq = task_rq_lock(p, &flags);
6161 p->cpus_allowed = cpus_allowed;
6162 dest_cpu = any_online_cpu(p->cpus_allowed);
6163 task_rq_unlock(rq, &flags);
6164
6165 /*
6166 * Don't tell them about moving exiting tasks or
6167 * kernel threads (both mm NULL), since they never
6168 * leave kernel.
6169 */
6170 if (p->mm && printk_ratelimit()) {
6171 printk(KERN_INFO "process %d (%s) no "
6172 "longer affine to cpu%d\n",
6173 task_pid_nr(p), p->comm, dead_cpu);
6174 }
6175 } 6169 }
6176 } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); 6170 }
6171
6172move:
6173 /* It can have affinity changed while we were choosing. */
6174 if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu)))
6175 goto again;
6177} 6176}
6178 6177
6179/* 6178/*
@@ -6185,7 +6184,7 @@ static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
6185 */ 6184 */
6186static void migrate_nr_uninterruptible(struct rq *rq_src) 6185static void migrate_nr_uninterruptible(struct rq *rq_src)
6187{ 6186{
6188 struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); 6187 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask));
6189 unsigned long flags; 6188 unsigned long flags;
6190 6189
6191 local_irq_save(flags); 6190 local_irq_save(flags);
@@ -6475,7 +6474,7 @@ static void set_rq_online(struct rq *rq)
6475 if (!rq->online) { 6474 if (!rq->online) {
6476 const struct sched_class *class; 6475 const struct sched_class *class;
6477 6476
6478 cpu_set(rq->cpu, rq->rd->online); 6477 cpumask_set_cpu(rq->cpu, rq->rd->online);
6479 rq->online = 1; 6478 rq->online = 1;
6480 6479
6481 for_each_class(class) { 6480 for_each_class(class) {
@@ -6495,7 +6494,7 @@ static void set_rq_offline(struct rq *rq)
6495 class->rq_offline(rq); 6494 class->rq_offline(rq);
6496 } 6495 }
6497 6496
6498 cpu_clear(rq->cpu, rq->rd->online); 6497 cpumask_clear_cpu(rq->cpu, rq->rd->online);
6499 rq->online = 0; 6498 rq->online = 0;
6500 } 6499 }
6501} 6500}
@@ -6536,7 +6535,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6536 rq = cpu_rq(cpu); 6535 rq = cpu_rq(cpu);
6537 spin_lock_irqsave(&rq->lock, flags); 6536 spin_lock_irqsave(&rq->lock, flags);
6538 if (rq->rd) { 6537 if (rq->rd) {
6539 BUG_ON(!cpu_isset(cpu, rq->rd->span)); 6538 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
6540 6539
6541 set_rq_online(rq); 6540 set_rq_online(rq);
6542 } 6541 }
@@ -6550,7 +6549,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6550 break; 6549 break;
6551 /* Unbind it from offline cpu so it can run. Fall thru. */ 6550 /* Unbind it from offline cpu so it can run. Fall thru. */
6552 kthread_bind(cpu_rq(cpu)->migration_thread, 6551 kthread_bind(cpu_rq(cpu)->migration_thread,
6553 any_online_cpu(cpu_online_map)); 6552 cpumask_any(cpu_online_mask));
6554 kthread_stop(cpu_rq(cpu)->migration_thread); 6553 kthread_stop(cpu_rq(cpu)->migration_thread);
6555 cpu_rq(cpu)->migration_thread = NULL; 6554 cpu_rq(cpu)->migration_thread = NULL;
6556 break; 6555 break;
@@ -6600,7 +6599,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6600 rq = cpu_rq(cpu); 6599 rq = cpu_rq(cpu);
6601 spin_lock_irqsave(&rq->lock, flags); 6600 spin_lock_irqsave(&rq->lock, flags);
6602 if (rq->rd) { 6601 if (rq->rd) {
6603 BUG_ON(!cpu_isset(cpu, rq->rd->span)); 6602 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
6604 set_rq_offline(rq); 6603 set_rq_offline(rq);
6605 } 6604 }
6606 spin_unlock_irqrestore(&rq->lock, flags); 6605 spin_unlock_irqrestore(&rq->lock, flags);
@@ -6638,36 +6637,14 @@ early_initcall(migration_init);
6638 6637
6639#ifdef CONFIG_SCHED_DEBUG 6638#ifdef CONFIG_SCHED_DEBUG
6640 6639
6641static inline const char *sd_level_to_string(enum sched_domain_level lvl)
6642{
6643 switch (lvl) {
6644 case SD_LV_NONE:
6645 return "NONE";
6646 case SD_LV_SIBLING:
6647 return "SIBLING";
6648 case SD_LV_MC:
6649 return "MC";
6650 case SD_LV_CPU:
6651 return "CPU";
6652 case SD_LV_NODE:
6653 return "NODE";
6654 case SD_LV_ALLNODES:
6655 return "ALLNODES";
6656 case SD_LV_MAX:
6657 return "MAX";
6658
6659 }
6660 return "MAX";
6661}
6662
6663static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, 6640static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6664 cpumask_t *groupmask) 6641 struct cpumask *groupmask)
6665{ 6642{
6666 struct sched_group *group = sd->groups; 6643 struct sched_group *group = sd->groups;
6667 char str[256]; 6644 char str[256];
6668 6645
6669 cpulist_scnprintf(str, sizeof(str), &sd->span); 6646 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
6670 cpus_clear(*groupmask); 6647 cpumask_clear(groupmask);
6671 6648
6672 printk(KERN_DEBUG "%*s domain %d: ", level, "", level); 6649 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6673 6650
@@ -6679,14 +6656,13 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6679 return -1; 6656 return -1;
6680 } 6657 }
6681 6658
6682 printk(KERN_CONT "span %s level %s\n", 6659 printk(KERN_CONT "span %s level %s\n", str, sd->name);
6683 str, sd_level_to_string(sd->level));
6684 6660
6685 if (!cpu_isset(cpu, sd->span)) { 6661 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
6686 printk(KERN_ERR "ERROR: domain->span does not contain " 6662 printk(KERN_ERR "ERROR: domain->span does not contain "
6687 "CPU%d\n", cpu); 6663 "CPU%d\n", cpu);
6688 } 6664 }
6689 if (!cpu_isset(cpu, group->cpumask)) { 6665 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
6690 printk(KERN_ERR "ERROR: domain->groups does not contain" 6666 printk(KERN_ERR "ERROR: domain->groups does not contain"
6691 " CPU%d\n", cpu); 6667 " CPU%d\n", cpu);
6692 } 6668 }
@@ -6706,31 +6682,32 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6706 break; 6682 break;
6707 } 6683 }
6708 6684
6709 if (!cpus_weight(group->cpumask)) { 6685 if (!cpumask_weight(sched_group_cpus(group))) {
6710 printk(KERN_CONT "\n"); 6686 printk(KERN_CONT "\n");
6711 printk(KERN_ERR "ERROR: empty group\n"); 6687 printk(KERN_ERR "ERROR: empty group\n");
6712 break; 6688 break;
6713 } 6689 }
6714 6690
6715 if (cpus_intersects(*groupmask, group->cpumask)) { 6691 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
6716 printk(KERN_CONT "\n"); 6692 printk(KERN_CONT "\n");
6717 printk(KERN_ERR "ERROR: repeated CPUs\n"); 6693 printk(KERN_ERR "ERROR: repeated CPUs\n");
6718 break; 6694 break;
6719 } 6695 }
6720 6696
6721 cpus_or(*groupmask, *groupmask, group->cpumask); 6697 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
6722 6698
6723 cpulist_scnprintf(str, sizeof(str), &group->cpumask); 6699 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
6724 printk(KERN_CONT " %s", str); 6700 printk(KERN_CONT " %s", str);
6725 6701
6726 group = group->next; 6702 group = group->next;
6727 } while (group != sd->groups); 6703 } while (group != sd->groups);
6728 printk(KERN_CONT "\n"); 6704 printk(KERN_CONT "\n");
6729 6705
6730 if (!cpus_equal(sd->span, *groupmask)) 6706 if (!cpumask_equal(sched_domain_span(sd), groupmask))
6731 printk(KERN_ERR "ERROR: groups don't span domain->span\n"); 6707 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
6732 6708
6733 if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) 6709 if (sd->parent &&
6710 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
6734 printk(KERN_ERR "ERROR: parent span is not a superset " 6711 printk(KERN_ERR "ERROR: parent span is not a superset "
6735 "of domain->span\n"); 6712 "of domain->span\n");
6736 return 0; 6713 return 0;
@@ -6738,7 +6715,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6738 6715
6739static void sched_domain_debug(struct sched_domain *sd, int cpu) 6716static void sched_domain_debug(struct sched_domain *sd, int cpu)
6740{ 6717{
6741 cpumask_t *groupmask; 6718 cpumask_var_t groupmask;
6742 int level = 0; 6719 int level = 0;
6743 6720
6744 if (!sd) { 6721 if (!sd) {
@@ -6748,8 +6725,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
6748 6725
6749 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); 6726 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6750 6727
6751 groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); 6728 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
6752 if (!groupmask) {
6753 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); 6729 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6754 return; 6730 return;
6755 } 6731 }
@@ -6762,7 +6738,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
6762 if (!sd) 6738 if (!sd)
6763 break; 6739 break;
6764 } 6740 }
6765 kfree(groupmask); 6741 free_cpumask_var(groupmask);
6766} 6742}
6767#else /* !CONFIG_SCHED_DEBUG */ 6743#else /* !CONFIG_SCHED_DEBUG */
6768# define sched_domain_debug(sd, cpu) do { } while (0) 6744# define sched_domain_debug(sd, cpu) do { } while (0)
@@ -6770,7 +6746,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
6770 6746
6771static int sd_degenerate(struct sched_domain *sd) 6747static int sd_degenerate(struct sched_domain *sd)
6772{ 6748{
6773 if (cpus_weight(sd->span) == 1) 6749 if (cpumask_weight(sched_domain_span(sd)) == 1)
6774 return 1; 6750 return 1;
6775 6751
6776 /* Following flags need at least 2 groups */ 6752 /* Following flags need at least 2 groups */
@@ -6801,7 +6777,7 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6801 if (sd_degenerate(parent)) 6777 if (sd_degenerate(parent))
6802 return 1; 6778 return 1;
6803 6779
6804 if (!cpus_equal(sd->span, parent->span)) 6780 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
6805 return 0; 6781 return 0;
6806 6782
6807 /* Does parent contain flags not in child? */ 6783 /* Does parent contain flags not in child? */
@@ -6816,6 +6792,8 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6816 SD_BALANCE_EXEC | 6792 SD_BALANCE_EXEC |
6817 SD_SHARE_CPUPOWER | 6793 SD_SHARE_CPUPOWER |
6818 SD_SHARE_PKG_RESOURCES); 6794 SD_SHARE_PKG_RESOURCES);
6795 if (nr_node_ids == 1)
6796 pflags &= ~SD_SERIALIZE;
6819 } 6797 }
6820 if (~cflags & pflags) 6798 if (~cflags & pflags)
6821 return 0; 6799 return 0;
@@ -6823,6 +6801,16 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6823 return 1; 6801 return 1;
6824} 6802}
6825 6803
6804static void free_rootdomain(struct root_domain *rd)
6805{
6806 cpupri_cleanup(&rd->cpupri);
6807
6808 free_cpumask_var(rd->rto_mask);
6809 free_cpumask_var(rd->online);
6810 free_cpumask_var(rd->span);
6811 kfree(rd);
6812}
6813
6826static void rq_attach_root(struct rq *rq, struct root_domain *rd) 6814static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6827{ 6815{
6828 unsigned long flags; 6816 unsigned long flags;
@@ -6832,38 +6820,63 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6832 if (rq->rd) { 6820 if (rq->rd) {
6833 struct root_domain *old_rd = rq->rd; 6821 struct root_domain *old_rd = rq->rd;
6834 6822
6835 if (cpu_isset(rq->cpu, old_rd->online)) 6823 if (cpumask_test_cpu(rq->cpu, old_rd->online))
6836 set_rq_offline(rq); 6824 set_rq_offline(rq);
6837 6825
6838 cpu_clear(rq->cpu, old_rd->span); 6826 cpumask_clear_cpu(rq->cpu, old_rd->span);
6839 6827
6840 if (atomic_dec_and_test(&old_rd->refcount)) 6828 if (atomic_dec_and_test(&old_rd->refcount))
6841 kfree(old_rd); 6829 free_rootdomain(old_rd);
6842 } 6830 }
6843 6831
6844 atomic_inc(&rd->refcount); 6832 atomic_inc(&rd->refcount);
6845 rq->rd = rd; 6833 rq->rd = rd;
6846 6834
6847 cpu_set(rq->cpu, rd->span); 6835 cpumask_set_cpu(rq->cpu, rd->span);
6848 if (cpu_isset(rq->cpu, cpu_online_map)) 6836 if (cpumask_test_cpu(rq->cpu, cpu_online_mask))
6849 set_rq_online(rq); 6837 set_rq_online(rq);
6850 6838
6851 spin_unlock_irqrestore(&rq->lock, flags); 6839 spin_unlock_irqrestore(&rq->lock, flags);
6852} 6840}
6853 6841
6854static void init_rootdomain(struct root_domain *rd) 6842static int init_rootdomain(struct root_domain *rd, bool bootmem)
6855{ 6843{
6856 memset(rd, 0, sizeof(*rd)); 6844 memset(rd, 0, sizeof(*rd));
6857 6845
6858 cpus_clear(rd->span); 6846 if (bootmem) {
6859 cpus_clear(rd->online); 6847 alloc_bootmem_cpumask_var(&def_root_domain.span);
6848 alloc_bootmem_cpumask_var(&def_root_domain.online);
6849 alloc_bootmem_cpumask_var(&def_root_domain.rto_mask);
6850 cpupri_init(&rd->cpupri, true);
6851 return 0;
6852 }
6853
6854 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
6855 goto free_rd;
6856 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
6857 goto free_span;
6858 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
6859 goto free_online;
6860
6861 if (cpupri_init(&rd->cpupri, false) != 0)
6862 goto free_rto_mask;
6863 return 0;
6860 6864
6861 cpupri_init(&rd->cpupri); 6865free_rto_mask:
6866 free_cpumask_var(rd->rto_mask);
6867free_online:
6868 free_cpumask_var(rd->online);
6869free_span:
6870 free_cpumask_var(rd->span);
6871free_rd:
6872 kfree(rd);
6873 return -ENOMEM;
6862} 6874}
6863 6875
6864static void init_defrootdomain(void) 6876static void init_defrootdomain(void)
6865{ 6877{
6866 init_rootdomain(&def_root_domain); 6878 init_rootdomain(&def_root_domain, true);
6879
6867 atomic_set(&def_root_domain.refcount, 1); 6880 atomic_set(&def_root_domain.refcount, 1);
6868} 6881}
6869 6882
@@ -6875,7 +6888,10 @@ static struct root_domain *alloc_rootdomain(void)
6875 if (!rd) 6888 if (!rd)
6876 return NULL; 6889 return NULL;
6877 6890
6878 init_rootdomain(rd); 6891 if (init_rootdomain(rd, false) != 0) {
6892 kfree(rd);
6893 return NULL;
6894 }
6879 6895
6880 return rd; 6896 return rd;
6881} 6897}
@@ -6917,19 +6933,12 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
6917} 6933}
6918 6934
6919/* cpus with isolated domains */ 6935/* cpus with isolated domains */
6920static cpumask_t cpu_isolated_map = CPU_MASK_NONE; 6936static cpumask_var_t cpu_isolated_map;
6921 6937
6922/* Setup the mask of cpus configured for isolated domains */ 6938/* Setup the mask of cpus configured for isolated domains */
6923static int __init isolated_cpu_setup(char *str) 6939static int __init isolated_cpu_setup(char *str)
6924{ 6940{
6925 static int __initdata ints[NR_CPUS]; 6941 cpulist_parse(str, cpu_isolated_map);
6926 int i;
6927
6928 str = get_options(str, ARRAY_SIZE(ints), ints);
6929 cpus_clear(cpu_isolated_map);
6930 for (i = 1; i <= ints[0]; i++)
6931 if (ints[i] < NR_CPUS)
6932 cpu_set(ints[i], cpu_isolated_map);
6933 return 1; 6942 return 1;
6934} 6943}
6935 6944
@@ -6938,42 +6947,43 @@ __setup("isolcpus=", isolated_cpu_setup);
6938/* 6947/*
6939 * init_sched_build_groups takes the cpumask we wish to span, and a pointer 6948 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6940 * to a function which identifies what group(along with sched group) a CPU 6949 * to a function which identifies what group(along with sched group) a CPU
6941 * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS 6950 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6942 * (due to the fact that we keep track of groups covered with a cpumask_t). 6951 * (due to the fact that we keep track of groups covered with a struct cpumask).
6943 * 6952 *
6944 * init_sched_build_groups will build a circular linked list of the groups 6953 * init_sched_build_groups will build a circular linked list of the groups
6945 * covered by the given span, and will set each group's ->cpumask correctly, 6954 * covered by the given span, and will set each group's ->cpumask correctly,
6946 * and ->cpu_power to 0. 6955 * and ->cpu_power to 0.
6947 */ 6956 */
6948static void 6957static void
6949init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, 6958init_sched_build_groups(const struct cpumask *span,
6950 int (*group_fn)(int cpu, const cpumask_t *cpu_map, 6959 const struct cpumask *cpu_map,
6960 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
6951 struct sched_group **sg, 6961 struct sched_group **sg,
6952 cpumask_t *tmpmask), 6962 struct cpumask *tmpmask),
6953 cpumask_t *covered, cpumask_t *tmpmask) 6963 struct cpumask *covered, struct cpumask *tmpmask)
6954{ 6964{
6955 struct sched_group *first = NULL, *last = NULL; 6965 struct sched_group *first = NULL, *last = NULL;
6956 int i; 6966 int i;
6957 6967
6958 cpus_clear(*covered); 6968 cpumask_clear(covered);
6959 6969
6960 for_each_cpu_mask_nr(i, *span) { 6970 for_each_cpu(i, span) {
6961 struct sched_group *sg; 6971 struct sched_group *sg;
6962 int group = group_fn(i, cpu_map, &sg, tmpmask); 6972 int group = group_fn(i, cpu_map, &sg, tmpmask);
6963 int j; 6973 int j;
6964 6974
6965 if (cpu_isset(i, *covered)) 6975 if (cpumask_test_cpu(i, covered))
6966 continue; 6976 continue;
6967 6977
6968 cpus_clear(sg->cpumask); 6978 cpumask_clear(sched_group_cpus(sg));
6969 sg->__cpu_power = 0; 6979 sg->__cpu_power = 0;
6970 6980
6971 for_each_cpu_mask_nr(j, *span) { 6981 for_each_cpu(j, span) {
6972 if (group_fn(j, cpu_map, NULL, tmpmask) != group) 6982 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
6973 continue; 6983 continue;
6974 6984
6975 cpu_set(j, *covered); 6985 cpumask_set_cpu(j, covered);
6976 cpu_set(j, sg->cpumask); 6986 cpumask_set_cpu(j, sched_group_cpus(sg));
6977 } 6987 }
6978 if (!first) 6988 if (!first)
6979 first = sg; 6989 first = sg;
@@ -7037,9 +7047,10 @@ static int find_next_best_node(int node, nodemask_t *used_nodes)
7037 * should be one that prevents unnecessary balancing, but also spreads tasks 7047 * should be one that prevents unnecessary balancing, but also spreads tasks
7038 * out optimally. 7048 * out optimally.
7039 */ 7049 */
7040static void sched_domain_node_span(int node, cpumask_t *span) 7050static void sched_domain_node_span(int node, struct cpumask *span)
7041{ 7051{
7042 nodemask_t used_nodes; 7052 nodemask_t used_nodes;
7053 /* FIXME: use cpumask_of_node() */
7043 node_to_cpumask_ptr(nodemask, node); 7054 node_to_cpumask_ptr(nodemask, node);
7044 int i; 7055 int i;
7045 7056
@@ -7061,18 +7072,33 @@ static void sched_domain_node_span(int node, cpumask_t *span)
7061int sched_smt_power_savings = 0, sched_mc_power_savings = 0; 7072int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
7062 7073
7063/* 7074/*
7075 * The cpus mask in sched_group and sched_domain hangs off the end.
7076 * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space
7077 * for nr_cpu_ids < CONFIG_NR_CPUS.
7078 */
7079struct static_sched_group {
7080 struct sched_group sg;
7081 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
7082};
7083
7084struct static_sched_domain {
7085 struct sched_domain sd;
7086 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
7087};
7088
7089/*
7064 * SMT sched-domains: 7090 * SMT sched-domains:
7065 */ 7091 */
7066#ifdef CONFIG_SCHED_SMT 7092#ifdef CONFIG_SCHED_SMT
7067static DEFINE_PER_CPU(struct sched_domain, cpu_domains); 7093static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
7068static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); 7094static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus);
7069 7095
7070static int 7096static int
7071cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7097cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
7072 cpumask_t *unused) 7098 struct sched_group **sg, struct cpumask *unused)
7073{ 7099{
7074 if (sg) 7100 if (sg)
7075 *sg = &per_cpu(sched_group_cpus, cpu); 7101 *sg = &per_cpu(sched_group_cpus, cpu).sg;
7076 return cpu; 7102 return cpu;
7077} 7103}
7078#endif /* CONFIG_SCHED_SMT */ 7104#endif /* CONFIG_SCHED_SMT */
@@ -7081,56 +7107,55 @@ cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg,
7081 * multi-core sched-domains: 7107 * multi-core sched-domains:
7082 */ 7108 */
7083#ifdef CONFIG_SCHED_MC 7109#ifdef CONFIG_SCHED_MC
7084static DEFINE_PER_CPU(struct sched_domain, core_domains); 7110static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
7085static DEFINE_PER_CPU(struct sched_group, sched_group_core); 7111static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
7086#endif /* CONFIG_SCHED_MC */ 7112#endif /* CONFIG_SCHED_MC */
7087 7113
7088#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) 7114#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
7089static int 7115static int
7090cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7116cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
7091 cpumask_t *mask) 7117 struct sched_group **sg, struct cpumask *mask)
7092{ 7118{
7093 int group; 7119 int group;
7094 7120
7095 *mask = per_cpu(cpu_sibling_map, cpu); 7121 cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map);
7096 cpus_and(*mask, *mask, *cpu_map); 7122 group = cpumask_first(mask);
7097 group = first_cpu(*mask);
7098 if (sg) 7123 if (sg)
7099 *sg = &per_cpu(sched_group_core, group); 7124 *sg = &per_cpu(sched_group_core, group).sg;
7100 return group; 7125 return group;
7101} 7126}
7102#elif defined(CONFIG_SCHED_MC) 7127#elif defined(CONFIG_SCHED_MC)
7103static int 7128static int
7104cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7129cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
7105 cpumask_t *unused) 7130 struct sched_group **sg, struct cpumask *unused)
7106{ 7131{
7107 if (sg) 7132 if (sg)
7108 *sg = &per_cpu(sched_group_core, cpu); 7133 *sg = &per_cpu(sched_group_core, cpu).sg;
7109 return cpu; 7134 return cpu;
7110} 7135}
7111#endif 7136#endif
7112 7137
7113static DEFINE_PER_CPU(struct sched_domain, phys_domains); 7138static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
7114static DEFINE_PER_CPU(struct sched_group, sched_group_phys); 7139static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
7115 7140
7116static int 7141static int
7117cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7142cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
7118 cpumask_t *mask) 7143 struct sched_group **sg, struct cpumask *mask)
7119{ 7144{
7120 int group; 7145 int group;
7121#ifdef CONFIG_SCHED_MC 7146#ifdef CONFIG_SCHED_MC
7147 /* FIXME: Use cpu_coregroup_mask. */
7122 *mask = cpu_coregroup_map(cpu); 7148 *mask = cpu_coregroup_map(cpu);
7123 cpus_and(*mask, *mask, *cpu_map); 7149 cpus_and(*mask, *mask, *cpu_map);
7124 group = first_cpu(*mask); 7150 group = cpumask_first(mask);
7125#elif defined(CONFIG_SCHED_SMT) 7151#elif defined(CONFIG_SCHED_SMT)
7126 *mask = per_cpu(cpu_sibling_map, cpu); 7152 cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map);
7127 cpus_and(*mask, *mask, *cpu_map); 7153 group = cpumask_first(mask);
7128 group = first_cpu(*mask);
7129#else 7154#else
7130 group = cpu; 7155 group = cpu;
7131#endif 7156#endif
7132 if (sg) 7157 if (sg)
7133 *sg = &per_cpu(sched_group_phys, group); 7158 *sg = &per_cpu(sched_group_phys, group).sg;
7134 return group; 7159 return group;
7135} 7160}
7136 7161
@@ -7144,19 +7169,21 @@ static DEFINE_PER_CPU(struct sched_domain, node_domains);
7144static struct sched_group ***sched_group_nodes_bycpu; 7169static struct sched_group ***sched_group_nodes_bycpu;
7145 7170
7146static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); 7171static DEFINE_PER_CPU(struct sched_domain, allnodes_domains);
7147static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); 7172static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
7148 7173
7149static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, 7174static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
7150 struct sched_group **sg, cpumask_t *nodemask) 7175 struct sched_group **sg,
7176 struct cpumask *nodemask)
7151{ 7177{
7152 int group; 7178 int group;
7179 /* FIXME: use cpumask_of_node */
7180 node_to_cpumask_ptr(pnodemask, cpu_to_node(cpu));
7153 7181
7154 *nodemask = node_to_cpumask(cpu_to_node(cpu)); 7182 cpumask_and(nodemask, pnodemask, cpu_map);
7155 cpus_and(*nodemask, *nodemask, *cpu_map); 7183 group = cpumask_first(nodemask);
7156 group = first_cpu(*nodemask);
7157 7184
7158 if (sg) 7185 if (sg)
7159 *sg = &per_cpu(sched_group_allnodes, group); 7186 *sg = &per_cpu(sched_group_allnodes, group).sg;
7160 return group; 7187 return group;
7161} 7188}
7162 7189
@@ -7168,11 +7195,11 @@ static void init_numa_sched_groups_power(struct sched_group *group_head)
7168 if (!sg) 7195 if (!sg)
7169 return; 7196 return;
7170 do { 7197 do {
7171 for_each_cpu_mask_nr(j, sg->cpumask) { 7198 for_each_cpu(j, sched_group_cpus(sg)) {
7172 struct sched_domain *sd; 7199 struct sched_domain *sd;
7173 7200
7174 sd = &per_cpu(phys_domains, j); 7201 sd = &per_cpu(phys_domains, j).sd;
7175 if (j != first_cpu(sd->groups->cpumask)) { 7202 if (j != cpumask_first(sched_group_cpus(sd->groups))) {
7176 /* 7203 /*
7177 * Only add "power" once for each 7204 * Only add "power" once for each
7178 * physical package. 7205 * physical package.
@@ -7189,11 +7216,12 @@ static void init_numa_sched_groups_power(struct sched_group *group_head)
7189 7216
7190#ifdef CONFIG_NUMA 7217#ifdef CONFIG_NUMA
7191/* Free memory allocated for various sched_group structures */ 7218/* Free memory allocated for various sched_group structures */
7192static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) 7219static void free_sched_groups(const struct cpumask *cpu_map,
7220 struct cpumask *nodemask)
7193{ 7221{
7194 int cpu, i; 7222 int cpu, i;
7195 7223
7196 for_each_cpu_mask_nr(cpu, *cpu_map) { 7224 for_each_cpu(cpu, cpu_map) {
7197 struct sched_group **sched_group_nodes 7225 struct sched_group **sched_group_nodes
7198 = sched_group_nodes_bycpu[cpu]; 7226 = sched_group_nodes_bycpu[cpu];
7199 7227
@@ -7202,10 +7230,11 @@ static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask)
7202 7230
7203 for (i = 0; i < nr_node_ids; i++) { 7231 for (i = 0; i < nr_node_ids; i++) {
7204 struct sched_group *oldsg, *sg = sched_group_nodes[i]; 7232 struct sched_group *oldsg, *sg = sched_group_nodes[i];
7233 /* FIXME: Use cpumask_of_node */
7234 node_to_cpumask_ptr(pnodemask, i);
7205 7235
7206 *nodemask = node_to_cpumask(i); 7236 cpus_and(*nodemask, *pnodemask, *cpu_map);
7207 cpus_and(*nodemask, *nodemask, *cpu_map); 7237 if (cpumask_empty(nodemask))
7208 if (cpus_empty(*nodemask))
7209 continue; 7238 continue;
7210 7239
7211 if (sg == NULL) 7240 if (sg == NULL)
@@ -7223,7 +7252,8 @@ next_sg:
7223 } 7252 }
7224} 7253}
7225#else /* !CONFIG_NUMA */ 7254#else /* !CONFIG_NUMA */
7226static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) 7255static void free_sched_groups(const struct cpumask *cpu_map,
7256 struct cpumask *nodemask)
7227{ 7257{
7228} 7258}
7229#endif /* CONFIG_NUMA */ 7259#endif /* CONFIG_NUMA */
@@ -7249,7 +7279,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
7249 7279
7250 WARN_ON(!sd || !sd->groups); 7280 WARN_ON(!sd || !sd->groups);
7251 7281
7252 if (cpu != first_cpu(sd->groups->cpumask)) 7282 if (cpu != cpumask_first(sched_group_cpus(sd->groups)))
7253 return; 7283 return;
7254 7284
7255 child = sd->child; 7285 child = sd->child;
@@ -7314,40 +7344,6 @@ SD_INIT_FUNC(CPU)
7314 SD_INIT_FUNC(MC) 7344 SD_INIT_FUNC(MC)
7315#endif 7345#endif
7316 7346
7317/*
7318 * To minimize stack usage kmalloc room for cpumasks and share the
7319 * space as the usage in build_sched_domains() dictates. Used only
7320 * if the amount of space is significant.
7321 */
7322struct allmasks {
7323 cpumask_t tmpmask; /* make this one first */
7324 union {
7325 cpumask_t nodemask;
7326 cpumask_t this_sibling_map;
7327 cpumask_t this_core_map;
7328 };
7329 cpumask_t send_covered;
7330
7331#ifdef CONFIG_NUMA
7332 cpumask_t domainspan;
7333 cpumask_t covered;
7334 cpumask_t notcovered;
7335#endif
7336};
7337
7338#if NR_CPUS > 128
7339#define SCHED_CPUMASK_ALLOC 1
7340#define SCHED_CPUMASK_FREE(v) kfree(v)
7341#define SCHED_CPUMASK_DECLARE(v) struct allmasks *v
7342#else
7343#define SCHED_CPUMASK_ALLOC 0
7344#define SCHED_CPUMASK_FREE(v)
7345#define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v
7346#endif
7347
7348#define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \
7349 ((unsigned long)(a) + offsetof(struct allmasks, v))
7350
7351static int default_relax_domain_level = -1; 7347static int default_relax_domain_level = -1;
7352 7348
7353static int __init setup_relax_domain_level(char *str) 7349static int __init setup_relax_domain_level(char *str)
@@ -7387,17 +7383,38 @@ static void set_domain_attribute(struct sched_domain *sd,
7387 * Build sched domains for a given set of cpus and attach the sched domains 7383 * Build sched domains for a given set of cpus and attach the sched domains
7388 * to the individual cpus 7384 * to the individual cpus
7389 */ 7385 */
7390static int __build_sched_domains(const cpumask_t *cpu_map, 7386static int __build_sched_domains(const struct cpumask *cpu_map,
7391 struct sched_domain_attr *attr) 7387 struct sched_domain_attr *attr)
7392{ 7388{
7393 int i; 7389 int i, err = -ENOMEM;
7394 struct root_domain *rd; 7390 struct root_domain *rd;
7395 SCHED_CPUMASK_DECLARE(allmasks); 7391 cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered,
7396 cpumask_t *tmpmask; 7392 tmpmask;
7397#ifdef CONFIG_NUMA 7393#ifdef CONFIG_NUMA
7394 cpumask_var_t domainspan, covered, notcovered;
7398 struct sched_group **sched_group_nodes = NULL; 7395 struct sched_group **sched_group_nodes = NULL;
7399 int sd_allnodes = 0; 7396 int sd_allnodes = 0;
7400 7397
7398 if (!alloc_cpumask_var(&domainspan, GFP_KERNEL))
7399 goto out;
7400 if (!alloc_cpumask_var(&covered, GFP_KERNEL))
7401 goto free_domainspan;
7402 if (!alloc_cpumask_var(&notcovered, GFP_KERNEL))
7403 goto free_covered;
7404#endif
7405
7406 if (!alloc_cpumask_var(&nodemask, GFP_KERNEL))
7407 goto free_notcovered;
7408 if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL))
7409 goto free_nodemask;
7410 if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL))
7411 goto free_this_sibling_map;
7412 if (!alloc_cpumask_var(&send_covered, GFP_KERNEL))
7413 goto free_this_core_map;
7414 if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL))
7415 goto free_send_covered;
7416
7417#ifdef CONFIG_NUMA
7401 /* 7418 /*
7402 * Allocate the per-node list of sched groups 7419 * Allocate the per-node list of sched groups
7403 */ 7420 */
@@ -7405,55 +7422,37 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7405 GFP_KERNEL); 7422 GFP_KERNEL);
7406 if (!sched_group_nodes) { 7423 if (!sched_group_nodes) {
7407 printk(KERN_WARNING "Can not alloc sched group node list\n"); 7424 printk(KERN_WARNING "Can not alloc sched group node list\n");
7408 return -ENOMEM; 7425 goto free_tmpmask;
7409 } 7426 }
7410#endif 7427#endif
7411 7428
7412 rd = alloc_rootdomain(); 7429 rd = alloc_rootdomain();
7413 if (!rd) { 7430 if (!rd) {
7414 printk(KERN_WARNING "Cannot alloc root domain\n"); 7431 printk(KERN_WARNING "Cannot alloc root domain\n");
7415#ifdef CONFIG_NUMA 7432 goto free_sched_groups;
7416 kfree(sched_group_nodes);
7417#endif
7418 return -ENOMEM;
7419 } 7433 }
7420 7434
7421#if SCHED_CPUMASK_ALLOC
7422 /* get space for all scratch cpumask variables */
7423 allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL);
7424 if (!allmasks) {
7425 printk(KERN_WARNING "Cannot alloc cpumask array\n");
7426 kfree(rd);
7427#ifdef CONFIG_NUMA 7435#ifdef CONFIG_NUMA
7428 kfree(sched_group_nodes); 7436 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes;
7429#endif
7430 return -ENOMEM;
7431 }
7432#endif
7433 tmpmask = (cpumask_t *)allmasks;
7434
7435
7436#ifdef CONFIG_NUMA
7437 sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes;
7438#endif 7437#endif
7439 7438
7440 /* 7439 /*
7441 * Set up domains for cpus specified by the cpu_map. 7440 * Set up domains for cpus specified by the cpu_map.
7442 */ 7441 */
7443 for_each_cpu_mask_nr(i, *cpu_map) { 7442 for_each_cpu(i, cpu_map) {
7444 struct sched_domain *sd = NULL, *p; 7443 struct sched_domain *sd = NULL, *p;
7445 SCHED_CPUMASK_VAR(nodemask, allmasks);
7446 7444
7445 /* FIXME: use cpumask_of_node */
7447 *nodemask = node_to_cpumask(cpu_to_node(i)); 7446 *nodemask = node_to_cpumask(cpu_to_node(i));
7448 cpus_and(*nodemask, *nodemask, *cpu_map); 7447 cpus_and(*nodemask, *nodemask, *cpu_map);
7449 7448
7450#ifdef CONFIG_NUMA 7449#ifdef CONFIG_NUMA
7451 if (cpus_weight(*cpu_map) > 7450 if (cpumask_weight(cpu_map) >
7452 SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { 7451 SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) {
7453 sd = &per_cpu(allnodes_domains, i); 7452 sd = &per_cpu(allnodes_domains, i);
7454 SD_INIT(sd, ALLNODES); 7453 SD_INIT(sd, ALLNODES);
7455 set_domain_attribute(sd, attr); 7454 set_domain_attribute(sd, attr);
7456 sd->span = *cpu_map; 7455 cpumask_copy(sched_domain_span(sd), cpu_map);
7457 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); 7456 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask);
7458 p = sd; 7457 p = sd;
7459 sd_allnodes = 1; 7458 sd_allnodes = 1;
@@ -7463,18 +7462,19 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7463 sd = &per_cpu(node_domains, i); 7462 sd = &per_cpu(node_domains, i);
7464 SD_INIT(sd, NODE); 7463 SD_INIT(sd, NODE);
7465 set_domain_attribute(sd, attr); 7464 set_domain_attribute(sd, attr);
7466 sched_domain_node_span(cpu_to_node(i), &sd->span); 7465 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
7467 sd->parent = p; 7466 sd->parent = p;
7468 if (p) 7467 if (p)
7469 p->child = sd; 7468 p->child = sd;
7470 cpus_and(sd->span, sd->span, *cpu_map); 7469 cpumask_and(sched_domain_span(sd),
7470 sched_domain_span(sd), cpu_map);
7471#endif 7471#endif
7472 7472
7473 p = sd; 7473 p = sd;
7474 sd = &per_cpu(phys_domains, i); 7474 sd = &per_cpu(phys_domains, i).sd;
7475 SD_INIT(sd, CPU); 7475 SD_INIT(sd, CPU);
7476 set_domain_attribute(sd, attr); 7476 set_domain_attribute(sd, attr);
7477 sd->span = *nodemask; 7477 cpumask_copy(sched_domain_span(sd), nodemask);
7478 sd->parent = p; 7478 sd->parent = p;
7479 if (p) 7479 if (p)
7480 p->child = sd; 7480 p->child = sd;
@@ -7482,11 +7482,12 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7482 7482
7483#ifdef CONFIG_SCHED_MC 7483#ifdef CONFIG_SCHED_MC
7484 p = sd; 7484 p = sd;
7485 sd = &per_cpu(core_domains, i); 7485 sd = &per_cpu(core_domains, i).sd;
7486 SD_INIT(sd, MC); 7486 SD_INIT(sd, MC);
7487 set_domain_attribute(sd, attr); 7487 set_domain_attribute(sd, attr);
7488 sd->span = cpu_coregroup_map(i); 7488 *sched_domain_span(sd) = cpu_coregroup_map(i);
7489 cpus_and(sd->span, sd->span, *cpu_map); 7489 cpumask_and(sched_domain_span(sd),
7490 sched_domain_span(sd), cpu_map);
7490 sd->parent = p; 7491 sd->parent = p;
7491 p->child = sd; 7492 p->child = sd;
7492 cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); 7493 cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask);
@@ -7494,11 +7495,11 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7494 7495
7495#ifdef CONFIG_SCHED_SMT 7496#ifdef CONFIG_SCHED_SMT
7496 p = sd; 7497 p = sd;
7497 sd = &per_cpu(cpu_domains, i); 7498 sd = &per_cpu(cpu_domains, i).sd;
7498 SD_INIT(sd, SIBLING); 7499 SD_INIT(sd, SIBLING);
7499 set_domain_attribute(sd, attr); 7500 set_domain_attribute(sd, attr);
7500 sd->span = per_cpu(cpu_sibling_map, i); 7501 cpumask_and(sched_domain_span(sd),
7501 cpus_and(sd->span, sd->span, *cpu_map); 7502 &per_cpu(cpu_sibling_map, i), cpu_map);
7502 sd->parent = p; 7503 sd->parent = p;
7503 p->child = sd; 7504 p->child = sd;
7504 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); 7505 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask);
@@ -7507,13 +7508,10 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7507 7508
7508#ifdef CONFIG_SCHED_SMT 7509#ifdef CONFIG_SCHED_SMT
7509 /* Set up CPU (sibling) groups */ 7510 /* Set up CPU (sibling) groups */
7510 for_each_cpu_mask_nr(i, *cpu_map) { 7511 for_each_cpu(i, cpu_map) {
7511 SCHED_CPUMASK_VAR(this_sibling_map, allmasks); 7512 cpumask_and(this_sibling_map,
7512 SCHED_CPUMASK_VAR(send_covered, allmasks); 7513 &per_cpu(cpu_sibling_map, i), cpu_map);
7513 7514 if (i != cpumask_first(this_sibling_map))
7514 *this_sibling_map = per_cpu(cpu_sibling_map, i);
7515 cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map);
7516 if (i != first_cpu(*this_sibling_map))
7517 continue; 7515 continue;
7518 7516
7519 init_sched_build_groups(this_sibling_map, cpu_map, 7517 init_sched_build_groups(this_sibling_map, cpu_map,
@@ -7524,13 +7522,11 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7524 7522
7525#ifdef CONFIG_SCHED_MC 7523#ifdef CONFIG_SCHED_MC
7526 /* Set up multi-core groups */ 7524 /* Set up multi-core groups */
7527 for_each_cpu_mask_nr(i, *cpu_map) { 7525 for_each_cpu(i, cpu_map) {
7528 SCHED_CPUMASK_VAR(this_core_map, allmasks); 7526 /* FIXME: Use cpu_coregroup_mask */
7529 SCHED_CPUMASK_VAR(send_covered, allmasks);
7530
7531 *this_core_map = cpu_coregroup_map(i); 7527 *this_core_map = cpu_coregroup_map(i);
7532 cpus_and(*this_core_map, *this_core_map, *cpu_map); 7528 cpus_and(*this_core_map, *this_core_map, *cpu_map);
7533 if (i != first_cpu(*this_core_map)) 7529 if (i != cpumask_first(this_core_map))
7534 continue; 7530 continue;
7535 7531
7536 init_sched_build_groups(this_core_map, cpu_map, 7532 init_sched_build_groups(this_core_map, cpu_map,
@@ -7541,12 +7537,10 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7541 7537
7542 /* Set up physical groups */ 7538 /* Set up physical groups */
7543 for (i = 0; i < nr_node_ids; i++) { 7539 for (i = 0; i < nr_node_ids; i++) {
7544 SCHED_CPUMASK_VAR(nodemask, allmasks); 7540 /* FIXME: Use cpumask_of_node */
7545 SCHED_CPUMASK_VAR(send_covered, allmasks);
7546
7547 *nodemask = node_to_cpumask(i); 7541 *nodemask = node_to_cpumask(i);
7548 cpus_and(*nodemask, *nodemask, *cpu_map); 7542 cpus_and(*nodemask, *nodemask, *cpu_map);
7549 if (cpus_empty(*nodemask)) 7543 if (cpumask_empty(nodemask))
7550 continue; 7544 continue;
7551 7545
7552 init_sched_build_groups(nodemask, cpu_map, 7546 init_sched_build_groups(nodemask, cpu_map,
@@ -7557,8 +7551,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7557#ifdef CONFIG_NUMA 7551#ifdef CONFIG_NUMA
7558 /* Set up node groups */ 7552 /* Set up node groups */
7559 if (sd_allnodes) { 7553 if (sd_allnodes) {
7560 SCHED_CPUMASK_VAR(send_covered, allmasks);
7561
7562 init_sched_build_groups(cpu_map, cpu_map, 7554 init_sched_build_groups(cpu_map, cpu_map,
7563 &cpu_to_allnodes_group, 7555 &cpu_to_allnodes_group,
7564 send_covered, tmpmask); 7556 send_covered, tmpmask);
@@ -7567,58 +7559,58 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7567 for (i = 0; i < nr_node_ids; i++) { 7559 for (i = 0; i < nr_node_ids; i++) {
7568 /* Set up node groups */ 7560 /* Set up node groups */
7569 struct sched_group *sg, *prev; 7561 struct sched_group *sg, *prev;
7570 SCHED_CPUMASK_VAR(nodemask, allmasks);
7571 SCHED_CPUMASK_VAR(domainspan, allmasks);
7572 SCHED_CPUMASK_VAR(covered, allmasks);
7573 int j; 7562 int j;
7574 7563
7564 /* FIXME: Use cpumask_of_node */
7575 *nodemask = node_to_cpumask(i); 7565 *nodemask = node_to_cpumask(i);
7576 cpus_clear(*covered); 7566 cpumask_clear(covered);
7577 7567
7578 cpus_and(*nodemask, *nodemask, *cpu_map); 7568 cpus_and(*nodemask, *nodemask, *cpu_map);
7579 if (cpus_empty(*nodemask)) { 7569 if (cpumask_empty(nodemask)) {
7580 sched_group_nodes[i] = NULL; 7570 sched_group_nodes[i] = NULL;
7581 continue; 7571 continue;
7582 } 7572 }
7583 7573
7584 sched_domain_node_span(i, domainspan); 7574 sched_domain_node_span(i, domainspan);
7585 cpus_and(*domainspan, *domainspan, *cpu_map); 7575 cpumask_and(domainspan, domainspan, cpu_map);
7586 7576
7587 sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); 7577 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
7578 GFP_KERNEL, i);
7588 if (!sg) { 7579 if (!sg) {
7589 printk(KERN_WARNING "Can not alloc domain group for " 7580 printk(KERN_WARNING "Can not alloc domain group for "
7590 "node %d\n", i); 7581 "node %d\n", i);
7591 goto error; 7582 goto error;
7592 } 7583 }
7593 sched_group_nodes[i] = sg; 7584 sched_group_nodes[i] = sg;
7594 for_each_cpu_mask_nr(j, *nodemask) { 7585 for_each_cpu(j, nodemask) {
7595 struct sched_domain *sd; 7586 struct sched_domain *sd;
7596 7587
7597 sd = &per_cpu(node_domains, j); 7588 sd = &per_cpu(node_domains, j);
7598 sd->groups = sg; 7589 sd->groups = sg;
7599 } 7590 }
7600 sg->__cpu_power = 0; 7591 sg->__cpu_power = 0;
7601 sg->cpumask = *nodemask; 7592 cpumask_copy(sched_group_cpus(sg), nodemask);
7602 sg->next = sg; 7593 sg->next = sg;
7603 cpus_or(*covered, *covered, *nodemask); 7594 cpumask_or(covered, covered, nodemask);
7604 prev = sg; 7595 prev = sg;
7605 7596
7606 for (j = 0; j < nr_node_ids; j++) { 7597 for (j = 0; j < nr_node_ids; j++) {
7607 SCHED_CPUMASK_VAR(notcovered, allmasks);
7608 int n = (i + j) % nr_node_ids; 7598 int n = (i + j) % nr_node_ids;
7599 /* FIXME: Use cpumask_of_node */
7609 node_to_cpumask_ptr(pnodemask, n); 7600 node_to_cpumask_ptr(pnodemask, n);
7610 7601
7611 cpus_complement(*notcovered, *covered); 7602 cpumask_complement(notcovered, covered);
7612 cpus_and(*tmpmask, *notcovered, *cpu_map); 7603 cpumask_and(tmpmask, notcovered, cpu_map);
7613 cpus_and(*tmpmask, *tmpmask, *domainspan); 7604 cpumask_and(tmpmask, tmpmask, domainspan);
7614 if (cpus_empty(*tmpmask)) 7605 if (cpumask_empty(tmpmask))
7615 break; 7606 break;
7616 7607
7617 cpus_and(*tmpmask, *tmpmask, *pnodemask); 7608 cpumask_and(tmpmask, tmpmask, pnodemask);
7618 if (cpus_empty(*tmpmask)) 7609 if (cpumask_empty(tmpmask))
7619 continue; 7610 continue;
7620 7611
7621 sg = kmalloc_node(sizeof(struct sched_group), 7612 sg = kmalloc_node(sizeof(struct sched_group) +
7613 cpumask_size(),
7622 GFP_KERNEL, i); 7614 GFP_KERNEL, i);
7623 if (!sg) { 7615 if (!sg) {
7624 printk(KERN_WARNING 7616 printk(KERN_WARNING
@@ -7626,9 +7618,9 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7626 goto error; 7618 goto error;
7627 } 7619 }
7628 sg->__cpu_power = 0; 7620 sg->__cpu_power = 0;
7629 sg->cpumask = *tmpmask; 7621 cpumask_copy(sched_group_cpus(sg), tmpmask);
7630 sg->next = prev->next; 7622 sg->next = prev->next;
7631 cpus_or(*covered, *covered, *tmpmask); 7623 cpumask_or(covered, covered, tmpmask);
7632 prev->next = sg; 7624 prev->next = sg;
7633 prev = sg; 7625 prev = sg;
7634 } 7626 }
@@ -7637,22 +7629,22 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7637 7629
7638 /* Calculate CPU power for physical packages and nodes */ 7630 /* Calculate CPU power for physical packages and nodes */
7639#ifdef CONFIG_SCHED_SMT 7631#ifdef CONFIG_SCHED_SMT
7640 for_each_cpu_mask_nr(i, *cpu_map) { 7632 for_each_cpu(i, cpu_map) {
7641 struct sched_domain *sd = &per_cpu(cpu_domains, i); 7633 struct sched_domain *sd = &per_cpu(cpu_domains, i).sd;
7642 7634
7643 init_sched_groups_power(i, sd); 7635 init_sched_groups_power(i, sd);
7644 } 7636 }
7645#endif 7637#endif
7646#ifdef CONFIG_SCHED_MC 7638#ifdef CONFIG_SCHED_MC
7647 for_each_cpu_mask_nr(i, *cpu_map) { 7639 for_each_cpu(i, cpu_map) {
7648 struct sched_domain *sd = &per_cpu(core_domains, i); 7640 struct sched_domain *sd = &per_cpu(core_domains, i).sd;
7649 7641
7650 init_sched_groups_power(i, sd); 7642 init_sched_groups_power(i, sd);
7651 } 7643 }
7652#endif 7644#endif
7653 7645
7654 for_each_cpu_mask_nr(i, *cpu_map) { 7646 for_each_cpu(i, cpu_map) {
7655 struct sched_domain *sd = &per_cpu(phys_domains, i); 7647 struct sched_domain *sd = &per_cpu(phys_domains, i).sd;
7656 7648
7657 init_sched_groups_power(i, sd); 7649 init_sched_groups_power(i, sd);
7658 } 7650 }
@@ -7664,56 +7656,87 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7664 if (sd_allnodes) { 7656 if (sd_allnodes) {
7665 struct sched_group *sg; 7657 struct sched_group *sg;
7666 7658
7667 cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, 7659 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
7668 tmpmask); 7660 tmpmask);
7669 init_numa_sched_groups_power(sg); 7661 init_numa_sched_groups_power(sg);
7670 } 7662 }
7671#endif 7663#endif
7672 7664
7673 /* Attach the domains */ 7665 /* Attach the domains */
7674 for_each_cpu_mask_nr(i, *cpu_map) { 7666 for_each_cpu(i, cpu_map) {
7675 struct sched_domain *sd; 7667 struct sched_domain *sd;
7676#ifdef CONFIG_SCHED_SMT 7668#ifdef CONFIG_SCHED_SMT
7677 sd = &per_cpu(cpu_domains, i); 7669 sd = &per_cpu(cpu_domains, i).sd;
7678#elif defined(CONFIG_SCHED_MC) 7670#elif defined(CONFIG_SCHED_MC)
7679 sd = &per_cpu(core_domains, i); 7671 sd = &per_cpu(core_domains, i).sd;
7680#else 7672#else
7681 sd = &per_cpu(phys_domains, i); 7673 sd = &per_cpu(phys_domains, i).sd;
7682#endif 7674#endif
7683 cpu_attach_domain(sd, rd, i); 7675 cpu_attach_domain(sd, rd, i);
7684 } 7676 }
7685 7677
7686 SCHED_CPUMASK_FREE((void *)allmasks); 7678 err = 0;
7687 return 0; 7679
7680free_tmpmask:
7681 free_cpumask_var(tmpmask);
7682free_send_covered:
7683 free_cpumask_var(send_covered);
7684free_this_core_map:
7685 free_cpumask_var(this_core_map);
7686free_this_sibling_map:
7687 free_cpumask_var(this_sibling_map);
7688free_nodemask:
7689 free_cpumask_var(nodemask);
7690free_notcovered:
7691#ifdef CONFIG_NUMA
7692 free_cpumask_var(notcovered);
7693free_covered:
7694 free_cpumask_var(covered);
7695free_domainspan:
7696 free_cpumask_var(domainspan);
7697out:
7698#endif
7699 return err;
7700
7701free_sched_groups:
7702#ifdef CONFIG_NUMA
7703 kfree(sched_group_nodes);
7704#endif
7705 goto free_tmpmask;
7688 7706
7689#ifdef CONFIG_NUMA 7707#ifdef CONFIG_NUMA
7690error: 7708error:
7691 free_sched_groups(cpu_map, tmpmask); 7709 free_sched_groups(cpu_map, tmpmask);
7692 SCHED_CPUMASK_FREE((void *)allmasks); 7710 free_rootdomain(rd);
7693 kfree(rd); 7711 goto free_tmpmask;
7694 return -ENOMEM;
7695#endif 7712#endif
7696} 7713}
7697 7714
7698static int build_sched_domains(const cpumask_t *cpu_map) 7715static int build_sched_domains(const struct cpumask *cpu_map)
7699{ 7716{
7700 return __build_sched_domains(cpu_map, NULL); 7717 return __build_sched_domains(cpu_map, NULL);
7701} 7718}
7702 7719
7703static cpumask_t *doms_cur; /* current sched domains */ 7720static struct cpumask *doms_cur; /* current sched domains */
7704static int ndoms_cur; /* number of sched domains in 'doms_cur' */ 7721static int ndoms_cur; /* number of sched domains in 'doms_cur' */
7705static struct sched_domain_attr *dattr_cur; 7722static struct sched_domain_attr *dattr_cur;
7706 /* attribues of custom domains in 'doms_cur' */ 7723 /* attribues of custom domains in 'doms_cur' */
7707 7724
7708/* 7725/*
7709 * Special case: If a kmalloc of a doms_cur partition (array of 7726 * Special case: If a kmalloc of a doms_cur partition (array of
7710 * cpumask_t) fails, then fallback to a single sched domain, 7727 * cpumask) fails, then fallback to a single sched domain,
7711 * as determined by the single cpumask_t fallback_doms. 7728 * as determined by the single cpumask fallback_doms.
7712 */ 7729 */
7713static cpumask_t fallback_doms; 7730static cpumask_var_t fallback_doms;
7714 7731
7715void __attribute__((weak)) arch_update_cpu_topology(void) 7732/*
7733 * arch_update_cpu_topology lets virtualized architectures update the
7734 * cpu core maps. It is supposed to return 1 if the topology changed
7735 * or 0 if it stayed the same.
7736 */
7737int __attribute__((weak)) arch_update_cpu_topology(void)
7716{ 7738{
7739 return 0;
7717} 7740}
7718 7741
7719/* 7742/*
@@ -7721,16 +7744,16 @@ void __attribute__((weak)) arch_update_cpu_topology(void)
7721 * For now this just excludes isolated cpus, but could be used to 7744 * For now this just excludes isolated cpus, but could be used to
7722 * exclude other special cases in the future. 7745 * exclude other special cases in the future.
7723 */ 7746 */
7724static int arch_init_sched_domains(const cpumask_t *cpu_map) 7747static int arch_init_sched_domains(const struct cpumask *cpu_map)
7725{ 7748{
7726 int err; 7749 int err;
7727 7750
7728 arch_update_cpu_topology(); 7751 arch_update_cpu_topology();
7729 ndoms_cur = 1; 7752 ndoms_cur = 1;
7730 doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); 7753 doms_cur = kmalloc(cpumask_size(), GFP_KERNEL);
7731 if (!doms_cur) 7754 if (!doms_cur)
7732 doms_cur = &fallback_doms; 7755 doms_cur = fallback_doms;
7733 cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); 7756 cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map);
7734 dattr_cur = NULL; 7757 dattr_cur = NULL;
7735 err = build_sched_domains(doms_cur); 7758 err = build_sched_domains(doms_cur);
7736 register_sched_domain_sysctl(); 7759 register_sched_domain_sysctl();
@@ -7738,8 +7761,8 @@ static int arch_init_sched_domains(const cpumask_t *cpu_map)
7738 return err; 7761 return err;
7739} 7762}
7740 7763
7741static void arch_destroy_sched_domains(const cpumask_t *cpu_map, 7764static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7742 cpumask_t *tmpmask) 7765 struct cpumask *tmpmask)
7743{ 7766{
7744 free_sched_groups(cpu_map, tmpmask); 7767 free_sched_groups(cpu_map, tmpmask);
7745} 7768}
@@ -7748,17 +7771,16 @@ static void arch_destroy_sched_domains(const cpumask_t *cpu_map,
7748 * Detach sched domains from a group of cpus specified in cpu_map 7771 * Detach sched domains from a group of cpus specified in cpu_map
7749 * These cpus will now be attached to the NULL domain 7772 * These cpus will now be attached to the NULL domain
7750 */ 7773 */
7751static void detach_destroy_domains(const cpumask_t *cpu_map) 7774static void detach_destroy_domains(const struct cpumask *cpu_map)
7752{ 7775{
7753 cpumask_t tmpmask; 7776 /* Save because hotplug lock held. */
7777 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
7754 int i; 7778 int i;
7755 7779
7756 unregister_sched_domain_sysctl(); 7780 for_each_cpu(i, cpu_map)
7757
7758 for_each_cpu_mask_nr(i, *cpu_map)
7759 cpu_attach_domain(NULL, &def_root_domain, i); 7781 cpu_attach_domain(NULL, &def_root_domain, i);
7760 synchronize_sched(); 7782 synchronize_sched();
7761 arch_destroy_sched_domains(cpu_map, &tmpmask); 7783 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
7762} 7784}
7763 7785
7764/* handle null as "default" */ 7786/* handle null as "default" */
@@ -7783,7 +7805,7 @@ static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7783 * doms_new[] to the current sched domain partitioning, doms_cur[]. 7805 * doms_new[] to the current sched domain partitioning, doms_cur[].
7784 * It destroys each deleted domain and builds each new domain. 7806 * It destroys each deleted domain and builds each new domain.
7785 * 7807 *
7786 * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. 7808 * 'doms_new' is an array of cpumask's of length 'ndoms_new'.
7787 * The masks don't intersect (don't overlap.) We should setup one 7809 * The masks don't intersect (don't overlap.) We should setup one
7788 * sched domain for each mask. CPUs not in any of the cpumasks will 7810 * sched domain for each mask. CPUs not in any of the cpumasks will
7789 * not be load balanced. If the same cpumask appears both in the 7811 * not be load balanced. If the same cpumask appears both in the
@@ -7797,28 +7819,33 @@ static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7797 * the single partition 'fallback_doms', it also forces the domains 7819 * the single partition 'fallback_doms', it also forces the domains
7798 * to be rebuilt. 7820 * to be rebuilt.
7799 * 7821 *
7800 * If doms_new == NULL it will be replaced with cpu_online_map. 7822 * If doms_new == NULL it will be replaced with cpu_online_mask.
7801 * ndoms_new == 0 is a special case for destroying existing domains, 7823 * ndoms_new == 0 is a special case for destroying existing domains,
7802 * and it will not create the default domain. 7824 * and it will not create the default domain.
7803 * 7825 *
7804 * Call with hotplug lock held 7826 * Call with hotplug lock held
7805 */ 7827 */
7806void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, 7828/* FIXME: Change to struct cpumask *doms_new[] */
7829void partition_sched_domains(int ndoms_new, struct cpumask *doms_new,
7807 struct sched_domain_attr *dattr_new) 7830 struct sched_domain_attr *dattr_new)
7808{ 7831{
7809 int i, j, n; 7832 int i, j, n;
7833 int new_topology;
7810 7834
7811 mutex_lock(&sched_domains_mutex); 7835 mutex_lock(&sched_domains_mutex);
7812 7836
7813 /* always unregister in case we don't destroy any domains */ 7837 /* always unregister in case we don't destroy any domains */
7814 unregister_sched_domain_sysctl(); 7838 unregister_sched_domain_sysctl();
7815 7839
7840 /* Let architecture update cpu core mappings. */
7841 new_topology = arch_update_cpu_topology();
7842
7816 n = doms_new ? ndoms_new : 0; 7843 n = doms_new ? ndoms_new : 0;
7817 7844
7818 /* Destroy deleted domains */ 7845 /* Destroy deleted domains */
7819 for (i = 0; i < ndoms_cur; i++) { 7846 for (i = 0; i < ndoms_cur; i++) {
7820 for (j = 0; j < n; j++) { 7847 for (j = 0; j < n && !new_topology; j++) {
7821 if (cpus_equal(doms_cur[i], doms_new[j]) 7848 if (cpumask_equal(&doms_cur[i], &doms_new[j])
7822 && dattrs_equal(dattr_cur, i, dattr_new, j)) 7849 && dattrs_equal(dattr_cur, i, dattr_new, j))
7823 goto match1; 7850 goto match1;
7824 } 7851 }
@@ -7830,15 +7857,15 @@ match1:
7830 7857
7831 if (doms_new == NULL) { 7858 if (doms_new == NULL) {
7832 ndoms_cur = 0; 7859 ndoms_cur = 0;
7833 doms_new = &fallback_doms; 7860 doms_new = fallback_doms;
7834 cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); 7861 cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map);
7835 dattr_new = NULL; 7862 WARN_ON_ONCE(dattr_new);
7836 } 7863 }
7837 7864
7838 /* Build new domains */ 7865 /* Build new domains */
7839 for (i = 0; i < ndoms_new; i++) { 7866 for (i = 0; i < ndoms_new; i++) {
7840 for (j = 0; j < ndoms_cur; j++) { 7867 for (j = 0; j < ndoms_cur && !new_topology; j++) {
7841 if (cpus_equal(doms_new[i], doms_cur[j]) 7868 if (cpumask_equal(&doms_new[i], &doms_cur[j])
7842 && dattrs_equal(dattr_new, i, dattr_cur, j)) 7869 && dattrs_equal(dattr_new, i, dattr_cur, j))
7843 goto match2; 7870 goto match2;
7844 } 7871 }
@@ -7850,7 +7877,7 @@ match2:
7850 } 7877 }
7851 7878
7852 /* Remember the new sched domains */ 7879 /* Remember the new sched domains */
7853 if (doms_cur != &fallback_doms) 7880 if (doms_cur != fallback_doms)
7854 kfree(doms_cur); 7881 kfree(doms_cur);
7855 kfree(dattr_cur); /* kfree(NULL) is safe */ 7882 kfree(dattr_cur); /* kfree(NULL) is safe */
7856 doms_cur = doms_new; 7883 doms_cur = doms_new;
@@ -7990,7 +8017,9 @@ static int update_runtime(struct notifier_block *nfb,
7990 8017
7991void __init sched_init_smp(void) 8018void __init sched_init_smp(void)
7992{ 8019{
7993 cpumask_t non_isolated_cpus; 8020 cpumask_var_t non_isolated_cpus;
8021
8022 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7994 8023
7995#if defined(CONFIG_NUMA) 8024#if defined(CONFIG_NUMA)
7996 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), 8025 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
@@ -7999,10 +8028,10 @@ void __init sched_init_smp(void)
7999#endif 8028#endif
8000 get_online_cpus(); 8029 get_online_cpus();
8001 mutex_lock(&sched_domains_mutex); 8030 mutex_lock(&sched_domains_mutex);
8002 arch_init_sched_domains(&cpu_online_map); 8031 arch_init_sched_domains(cpu_online_mask);
8003 cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); 8032 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
8004 if (cpus_empty(non_isolated_cpus)) 8033 if (cpumask_empty(non_isolated_cpus))
8005 cpu_set(smp_processor_id(), non_isolated_cpus); 8034 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
8006 mutex_unlock(&sched_domains_mutex); 8035 mutex_unlock(&sched_domains_mutex);
8007 put_online_cpus(); 8036 put_online_cpus();
8008 8037
@@ -8017,9 +8046,13 @@ void __init sched_init_smp(void)
8017 init_hrtick(); 8046 init_hrtick();
8018 8047
8019 /* Move init over to a non-isolated CPU */ 8048 /* Move init over to a non-isolated CPU */
8020 if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) 8049 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
8021 BUG(); 8050 BUG();
8022 sched_init_granularity(); 8051 sched_init_granularity();
8052 free_cpumask_var(non_isolated_cpus);
8053
8054 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
8055 init_sched_rt_class();
8023} 8056}
8024#else 8057#else
8025void __init sched_init_smp(void) 8058void __init sched_init_smp(void)
@@ -8334,6 +8367,15 @@ void __init sched_init(void)
8334 */ 8367 */
8335 current->sched_class = &fair_sched_class; 8368 current->sched_class = &fair_sched_class;
8336 8369
8370 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
8371 alloc_bootmem_cpumask_var(&nohz_cpu_mask);
8372#ifdef CONFIG_SMP
8373#ifdef CONFIG_NO_HZ
8374 alloc_bootmem_cpumask_var(&nohz.cpu_mask);
8375#endif
8376 alloc_bootmem_cpumask_var(&cpu_isolated_map);
8377#endif /* SMP */
8378
8337 scheduler_running = 1; 8379 scheduler_running = 1;
8338} 8380}
8339 8381
@@ -8492,7 +8534,7 @@ static
8492int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) 8534int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8493{ 8535{
8494 struct cfs_rq *cfs_rq; 8536 struct cfs_rq *cfs_rq;
8495 struct sched_entity *se, *parent_se; 8537 struct sched_entity *se;
8496 struct rq *rq; 8538 struct rq *rq;
8497 int i; 8539 int i;
8498 8540
@@ -8508,18 +8550,17 @@ int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8508 for_each_possible_cpu(i) { 8550 for_each_possible_cpu(i) {
8509 rq = cpu_rq(i); 8551 rq = cpu_rq(i);
8510 8552
8511 cfs_rq = kmalloc_node(sizeof(struct cfs_rq), 8553 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8512 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8554 GFP_KERNEL, cpu_to_node(i));
8513 if (!cfs_rq) 8555 if (!cfs_rq)
8514 goto err; 8556 goto err;
8515 8557
8516 se = kmalloc_node(sizeof(struct sched_entity), 8558 se = kzalloc_node(sizeof(struct sched_entity),
8517 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8559 GFP_KERNEL, cpu_to_node(i));
8518 if (!se) 8560 if (!se)
8519 goto err; 8561 goto err;
8520 8562
8521 parent_se = parent ? parent->se[i] : NULL; 8563 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
8522 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se);
8523 } 8564 }
8524 8565
8525 return 1; 8566 return 1;
@@ -8580,7 +8621,7 @@ static
8580int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) 8621int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8581{ 8622{
8582 struct rt_rq *rt_rq; 8623 struct rt_rq *rt_rq;
8583 struct sched_rt_entity *rt_se, *parent_se; 8624 struct sched_rt_entity *rt_se;
8584 struct rq *rq; 8625 struct rq *rq;
8585 int i; 8626 int i;
8586 8627
@@ -8597,18 +8638,17 @@ int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8597 for_each_possible_cpu(i) { 8638 for_each_possible_cpu(i) {
8598 rq = cpu_rq(i); 8639 rq = cpu_rq(i);
8599 8640
8600 rt_rq = kmalloc_node(sizeof(struct rt_rq), 8641 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8601 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8642 GFP_KERNEL, cpu_to_node(i));
8602 if (!rt_rq) 8643 if (!rt_rq)
8603 goto err; 8644 goto err;
8604 8645
8605 rt_se = kmalloc_node(sizeof(struct sched_rt_entity), 8646 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8606 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8647 GFP_KERNEL, cpu_to_node(i));
8607 if (!rt_se) 8648 if (!rt_se)
8608 goto err; 8649 goto err;
8609 8650
8610 parent_se = parent ? parent->rt_se[i] : NULL; 8651 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
8611 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se);
8612 } 8652 }
8613 8653
8614 return 1; 8654 return 1;
@@ -9251,11 +9291,12 @@ struct cgroup_subsys cpu_cgroup_subsys = {
9251 * (balbir@in.ibm.com). 9291 * (balbir@in.ibm.com).
9252 */ 9292 */
9253 9293
9254/* track cpu usage of a group of tasks */ 9294/* track cpu usage of a group of tasks and its child groups */
9255struct cpuacct { 9295struct cpuacct {
9256 struct cgroup_subsys_state css; 9296 struct cgroup_subsys_state css;
9257 /* cpuusage holds pointer to a u64-type object on every cpu */ 9297 /* cpuusage holds pointer to a u64-type object on every cpu */
9258 u64 *cpuusage; 9298 u64 *cpuusage;
9299 struct cpuacct *parent;
9259}; 9300};
9260 9301
9261struct cgroup_subsys cpuacct_subsys; 9302struct cgroup_subsys cpuacct_subsys;
@@ -9289,6 +9330,9 @@ static struct cgroup_subsys_state *cpuacct_create(
9289 return ERR_PTR(-ENOMEM); 9330 return ERR_PTR(-ENOMEM);
9290 } 9331 }
9291 9332
9333 if (cgrp->parent)
9334 ca->parent = cgroup_ca(cgrp->parent);
9335
9292 return &ca->css; 9336 return &ca->css;
9293} 9337}
9294 9338
@@ -9368,14 +9412,16 @@ static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
9368static void cpuacct_charge(struct task_struct *tsk, u64 cputime) 9412static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9369{ 9413{
9370 struct cpuacct *ca; 9414 struct cpuacct *ca;
9415 int cpu;
9371 9416
9372 if (!cpuacct_subsys.active) 9417 if (!cpuacct_subsys.active)
9373 return; 9418 return;
9374 9419
9420 cpu = task_cpu(tsk);
9375 ca = task_ca(tsk); 9421 ca = task_ca(tsk);
9376 if (ca) {
9377 u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk));
9378 9422
9423 for (; ca; ca = ca->parent) {
9424 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu);
9379 *cpuusage += cputime; 9425 *cpuusage += cputime;
9380 } 9426 }
9381} 9427}