aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/sched.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/sched.c')
-rw-r--r--kernel/sched.c1468
1 files changed, 833 insertions, 635 deletions
diff --git a/kernel/sched.c b/kernel/sched.c
index b7480fb5c3dc..545c6fccd1dc 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)
@@ -203,7 +209,6 @@ void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
203 hrtimer_init(&rt_b->rt_period_timer, 209 hrtimer_init(&rt_b->rt_period_timer,
204 CLOCK_MONOTONIC, HRTIMER_MODE_REL); 210 CLOCK_MONOTONIC, HRTIMER_MODE_REL);
205 rt_b->rt_period_timer.function = sched_rt_period_timer; 211 rt_b->rt_period_timer.function = sched_rt_period_timer;
206 rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED;
207} 212}
208 213
209static inline int rt_bandwidth_enabled(void) 214static inline int rt_bandwidth_enabled(void)
@@ -261,6 +266,10 @@ struct task_group {
261 struct cgroup_subsys_state css; 266 struct cgroup_subsys_state css;
262#endif 267#endif
263 268
269#ifdef CONFIG_USER_SCHED
270 uid_t uid;
271#endif
272
264#ifdef CONFIG_FAIR_GROUP_SCHED 273#ifdef CONFIG_FAIR_GROUP_SCHED
265 /* schedulable entities of this group on each cpu */ 274 /* schedulable entities of this group on each cpu */
266 struct sched_entity **se; 275 struct sched_entity **se;
@@ -286,6 +295,12 @@ struct task_group {
286 295
287#ifdef CONFIG_USER_SCHED 296#ifdef CONFIG_USER_SCHED
288 297
298/* Helper function to pass uid information to create_sched_user() */
299void set_tg_uid(struct user_struct *user)
300{
301 user->tg->uid = user->uid;
302}
303
289/* 304/*
290 * Root task group. 305 * Root task group.
291 * Every UID task group (including init_task_group aka UID-0) will 306 * Every UID task group (including init_task_group aka UID-0) will
@@ -345,7 +360,9 @@ static inline struct task_group *task_group(struct task_struct *p)
345 struct task_group *tg; 360 struct task_group *tg;
346 361
347#ifdef CONFIG_USER_SCHED 362#ifdef CONFIG_USER_SCHED
348 tg = p->user->tg; 363 rcu_read_lock();
364 tg = __task_cred(p)->user->tg;
365 rcu_read_unlock();
349#elif defined(CONFIG_CGROUP_SCHED) 366#elif defined(CONFIG_CGROUP_SCHED)
350 tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), 367 tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id),
351 struct task_group, css); 368 struct task_group, css);
@@ -481,18 +498,26 @@ struct rt_rq {
481 */ 498 */
482struct root_domain { 499struct root_domain {
483 atomic_t refcount; 500 atomic_t refcount;
484 cpumask_t span; 501 cpumask_var_t span;
485 cpumask_t online; 502 cpumask_var_t online;
486 503
487 /* 504 /*
488 * The "RT overload" flag: it gets set if a CPU has more than 505 * The "RT overload" flag: it gets set if a CPU has more than
489 * one runnable RT task. 506 * one runnable RT task.
490 */ 507 */
491 cpumask_t rto_mask; 508 cpumask_var_t rto_mask;
492 atomic_t rto_count; 509 atomic_t rto_count;
493#ifdef CONFIG_SMP 510#ifdef CONFIG_SMP
494 struct cpupri cpupri; 511 struct cpupri cpupri;
495#endif 512#endif
513#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
514 /*
515 * Preferred wake up cpu nominated by sched_mc balance that will be
516 * used when most cpus are idle in the system indicating overall very
517 * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2)
518 */
519 unsigned int sched_mc_preferred_wakeup_cpu;
520#endif
496}; 521};
497 522
498/* 523/*
@@ -586,6 +611,8 @@ struct rq {
586#ifdef CONFIG_SCHEDSTATS 611#ifdef CONFIG_SCHEDSTATS
587 /* latency stats */ 612 /* latency stats */
588 struct sched_info rq_sched_info; 613 struct sched_info rq_sched_info;
614 unsigned long long rq_cpu_time;
615 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
589 616
590 /* sys_sched_yield() stats */ 617 /* sys_sched_yield() stats */
591 unsigned int yld_exp_empty; 618 unsigned int yld_exp_empty;
@@ -703,45 +730,18 @@ static __read_mostly char *sched_feat_names[] = {
703 730
704#undef SCHED_FEAT 731#undef SCHED_FEAT
705 732
706static int sched_feat_open(struct inode *inode, struct file *filp) 733static int sched_feat_show(struct seq_file *m, void *v)
707{ 734{
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; 735 int i;
720 736
721 for (i = 0; sched_feat_names[i]; i++) { 737 for (i = 0; sched_feat_names[i]; i++) {
722 len += strlen(sched_feat_names[i]); 738 if (!(sysctl_sched_features & (1UL << i)))
723 len += 4; 739 seq_puts(m, "NO_");
740 seq_printf(m, "%s ", sched_feat_names[i]);
724 } 741 }
742 seq_puts(m, "\n");
725 743
726 buf = kmalloc(len + 2, GFP_KERNEL); 744 return 0;
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 }
736
737 r += sprintf(buf + r, "\n");
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} 745}
746 746
747static ssize_t 747static ssize_t
@@ -786,10 +786,17 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
786 return cnt; 786 return cnt;
787} 787}
788 788
789static int sched_feat_open(struct inode *inode, struct file *filp)
790{
791 return single_open(filp, sched_feat_show, NULL);
792}
793
789static struct file_operations sched_feat_fops = { 794static struct file_operations sched_feat_fops = {
790 .open = sched_feat_open, 795 .open = sched_feat_open,
791 .read = sched_feat_read, 796 .write = sched_feat_write,
792 .write = sched_feat_write, 797 .read = seq_read,
798 .llseek = seq_lseek,
799 .release = single_release,
793}; 800};
794 801
795static __init int sched_init_debug(void) 802static __init int sched_init_debug(void)
@@ -1139,7 +1146,6 @@ static void init_rq_hrtick(struct rq *rq)
1139 1146
1140 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1147 hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1141 rq->hrtick_timer.function = hrtick; 1148 rq->hrtick_timer.function = hrtick;
1142 rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
1143} 1149}
1144#else /* CONFIG_SCHED_HRTICK */ 1150#else /* CONFIG_SCHED_HRTICK */
1145static inline void hrtick_clear(struct rq *rq) 1151static inline void hrtick_clear(struct rq *rq)
@@ -1474,27 +1480,13 @@ static void
1474update_group_shares_cpu(struct task_group *tg, int cpu, 1480update_group_shares_cpu(struct task_group *tg, int cpu,
1475 unsigned long sd_shares, unsigned long sd_rq_weight) 1481 unsigned long sd_shares, unsigned long sd_rq_weight)
1476{ 1482{
1477 int boost = 0;
1478 unsigned long shares; 1483 unsigned long shares;
1479 unsigned long rq_weight; 1484 unsigned long rq_weight;
1480 1485
1481 if (!tg->se[cpu]) 1486 if (!tg->se[cpu])
1482 return; 1487 return;
1483 1488
1484 rq_weight = tg->cfs_rq[cpu]->load.weight; 1489 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 1490
1499 /* 1491 /*
1500 * \Sum shares * rq_weight 1492 * \Sum shares * rq_weight
@@ -1502,7 +1494,7 @@ update_group_shares_cpu(struct task_group *tg, int cpu,
1502 * \Sum rq_weight 1494 * \Sum rq_weight
1503 * 1495 *
1504 */ 1496 */
1505 shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); 1497 shares = (sd_shares * rq_weight) / sd_rq_weight;
1506 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); 1498 shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES);
1507 1499
1508 if (abs(shares - tg->se[cpu]->load.weight) > 1500 if (abs(shares - tg->se[cpu]->load.weight) >
@@ -1511,11 +1503,7 @@ update_group_shares_cpu(struct task_group *tg, int cpu,
1511 unsigned long flags; 1503 unsigned long flags;
1512 1504
1513 spin_lock_irqsave(&rq->lock, flags); 1505 spin_lock_irqsave(&rq->lock, flags);
1514 /* 1506 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 1507
1520 __set_se_shares(tg->se[cpu], shares); 1508 __set_se_shares(tg->se[cpu], shares);
1521 spin_unlock_irqrestore(&rq->lock, flags); 1509 spin_unlock_irqrestore(&rq->lock, flags);
@@ -1529,13 +1517,23 @@ update_group_shares_cpu(struct task_group *tg, int cpu,
1529 */ 1517 */
1530static int tg_shares_up(struct task_group *tg, void *data) 1518static int tg_shares_up(struct task_group *tg, void *data)
1531{ 1519{
1532 unsigned long rq_weight = 0; 1520 unsigned long weight, rq_weight = 0;
1533 unsigned long shares = 0; 1521 unsigned long shares = 0;
1534 struct sched_domain *sd = data; 1522 struct sched_domain *sd = data;
1535 int i; 1523 int i;
1536 1524
1537 for_each_cpu_mask(i, sd->span) { 1525 for_each_cpu(i, sched_domain_span(sd)) {
1538 rq_weight += tg->cfs_rq[i]->load.weight; 1526 /*
1527 * If there are currently no tasks on the cpu pretend there
1528 * is one of average load so that when a new task gets to
1529 * run here it will not get delayed by group starvation.
1530 */
1531 weight = tg->cfs_rq[i]->load.weight;
1532 if (!weight)
1533 weight = NICE_0_LOAD;
1534
1535 tg->cfs_rq[i]->rq_weight = weight;
1536 rq_weight += weight;
1539 shares += tg->cfs_rq[i]->shares; 1537 shares += tg->cfs_rq[i]->shares;
1540 } 1538 }
1541 1539
@@ -1545,10 +1543,7 @@ static int tg_shares_up(struct task_group *tg, void *data)
1545 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) 1543 if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE))
1546 shares = tg->shares; 1544 shares = tg->shares;
1547 1545
1548 if (!rq_weight) 1546 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); 1547 update_group_shares_cpu(tg, i, shares, rq_weight);
1553 1548
1554 return 0; 1549 return 0;
@@ -1612,6 +1607,39 @@ static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd)
1612 1607
1613#endif 1608#endif
1614 1609
1610/*
1611 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1612 */
1613static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1614 __releases(this_rq->lock)
1615 __acquires(busiest->lock)
1616 __acquires(this_rq->lock)
1617{
1618 int ret = 0;
1619
1620 if (unlikely(!irqs_disabled())) {
1621 /* printk() doesn't work good under rq->lock */
1622 spin_unlock(&this_rq->lock);
1623 BUG_ON(1);
1624 }
1625 if (unlikely(!spin_trylock(&busiest->lock))) {
1626 if (busiest < this_rq) {
1627 spin_unlock(&this_rq->lock);
1628 spin_lock(&busiest->lock);
1629 spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING);
1630 ret = 1;
1631 } else
1632 spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING);
1633 }
1634 return ret;
1635}
1636
1637static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1638 __releases(busiest->lock)
1639{
1640 spin_unlock(&busiest->lock);
1641 lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_);
1642}
1615#endif 1643#endif
1616 1644
1617#ifdef CONFIG_FAIR_GROUP_SCHED 1645#ifdef CONFIG_FAIR_GROUP_SCHED
@@ -1845,6 +1873,8 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
1845 1873
1846 clock_offset = old_rq->clock - new_rq->clock; 1874 clock_offset = old_rq->clock - new_rq->clock;
1847 1875
1876 trace_sched_migrate_task(p, task_cpu(p), new_cpu);
1877
1848#ifdef CONFIG_SCHEDSTATS 1878#ifdef CONFIG_SCHEDSTATS
1849 if (p->se.wait_start) 1879 if (p->se.wait_start)
1850 p->se.wait_start -= clock_offset; 1880 p->se.wait_start -= clock_offset;
@@ -2079,15 +2109,17 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu)
2079 int i; 2109 int i;
2080 2110
2081 /* Skip over this group if it has no CPUs allowed */ 2111 /* Skip over this group if it has no CPUs allowed */
2082 if (!cpus_intersects(group->cpumask, p->cpus_allowed)) 2112 if (!cpumask_intersects(sched_group_cpus(group),
2113 &p->cpus_allowed))
2083 continue; 2114 continue;
2084 2115
2085 local_group = cpu_isset(this_cpu, group->cpumask); 2116 local_group = cpumask_test_cpu(this_cpu,
2117 sched_group_cpus(group));
2086 2118
2087 /* Tally up the load of all CPUs in the group */ 2119 /* Tally up the load of all CPUs in the group */
2088 avg_load = 0; 2120 avg_load = 0;
2089 2121
2090 for_each_cpu_mask_nr(i, group->cpumask) { 2122 for_each_cpu(i, sched_group_cpus(group)) {
2091 /* Bias balancing toward cpus of our domain */ 2123 /* Bias balancing toward cpus of our domain */
2092 if (local_group) 2124 if (local_group)
2093 load = source_load(i, load_idx); 2125 load = source_load(i, load_idx);
@@ -2119,17 +2151,14 @@ 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. 2151 * find_idlest_cpu - find the idlest cpu among the cpus in group.
2120 */ 2152 */
2121static int 2153static int
2122find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, 2154find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
2123 cpumask_t *tmp)
2124{ 2155{
2125 unsigned long load, min_load = ULONG_MAX; 2156 unsigned long load, min_load = ULONG_MAX;
2126 int idlest = -1; 2157 int idlest = -1;
2127 int i; 2158 int i;
2128 2159
2129 /* Traverse only the allowed CPUs */ 2160 /* Traverse only the allowed CPUs */
2130 cpus_and(*tmp, group->cpumask, p->cpus_allowed); 2161 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); 2162 load = weighted_cpuload(i);
2134 2163
2135 if (load < min_load || (load == min_load && i == this_cpu)) { 2164 if (load < min_load || (load == min_load && i == this_cpu)) {
@@ -2171,7 +2200,6 @@ static int sched_balance_self(int cpu, int flag)
2171 update_shares(sd); 2200 update_shares(sd);
2172 2201
2173 while (sd) { 2202 while (sd) {
2174 cpumask_t span, tmpmask;
2175 struct sched_group *group; 2203 struct sched_group *group;
2176 int new_cpu, weight; 2204 int new_cpu, weight;
2177 2205
@@ -2180,14 +2208,13 @@ static int sched_balance_self(int cpu, int flag)
2180 continue; 2208 continue;
2181 } 2209 }
2182 2210
2183 span = sd->span;
2184 group = find_idlest_group(sd, t, cpu); 2211 group = find_idlest_group(sd, t, cpu);
2185 if (!group) { 2212 if (!group) {
2186 sd = sd->child; 2213 sd = sd->child;
2187 continue; 2214 continue;
2188 } 2215 }
2189 2216
2190 new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); 2217 new_cpu = find_idlest_cpu(group, t, cpu);
2191 if (new_cpu == -1 || new_cpu == cpu) { 2218 if (new_cpu == -1 || new_cpu == cpu) {
2192 /* Now try balancing at a lower domain level of cpu */ 2219 /* Now try balancing at a lower domain level of cpu */
2193 sd = sd->child; 2220 sd = sd->child;
@@ -2196,10 +2223,10 @@ static int sched_balance_self(int cpu, int flag)
2196 2223
2197 /* Now try balancing at a lower domain level of new_cpu */ 2224 /* Now try balancing at a lower domain level of new_cpu */
2198 cpu = new_cpu; 2225 cpu = new_cpu;
2226 weight = cpumask_weight(sched_domain_span(sd));
2199 sd = NULL; 2227 sd = NULL;
2200 weight = cpus_weight(span);
2201 for_each_domain(cpu, tmp) { 2228 for_each_domain(cpu, tmp) {
2202 if (weight <= cpus_weight(tmp->span)) 2229 if (weight <= cpumask_weight(sched_domain_span(tmp)))
2203 break; 2230 break;
2204 if (tmp->flags & flag) 2231 if (tmp->flags & flag)
2205 sd = tmp; 2232 sd = tmp;
@@ -2244,7 +2271,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2244 cpu = task_cpu(p); 2271 cpu = task_cpu(p);
2245 2272
2246 for_each_domain(this_cpu, sd) { 2273 for_each_domain(this_cpu, sd) {
2247 if (cpu_isset(cpu, sd->span)) { 2274 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2248 update_shares(sd); 2275 update_shares(sd);
2249 break; 2276 break;
2250 } 2277 }
@@ -2254,6 +2281,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2254 2281
2255 smp_wmb(); 2282 smp_wmb();
2256 rq = task_rq_lock(p, &flags); 2283 rq = task_rq_lock(p, &flags);
2284 update_rq_clock(rq);
2257 old_state = p->state; 2285 old_state = p->state;
2258 if (!(old_state & state)) 2286 if (!(old_state & state))
2259 goto out; 2287 goto out;
@@ -2292,7 +2320,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2292 else { 2320 else {
2293 struct sched_domain *sd; 2321 struct sched_domain *sd;
2294 for_each_domain(this_cpu, sd) { 2322 for_each_domain(this_cpu, sd) {
2295 if (cpu_isset(cpu, sd->span)) { 2323 if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
2296 schedstat_inc(sd, ttwu_wake_remote); 2324 schedstat_inc(sd, ttwu_wake_remote);
2297 break; 2325 break;
2298 } 2326 }
@@ -2311,12 +2339,11 @@ out_activate:
2311 schedstat_inc(p, se.nr_wakeups_local); 2339 schedstat_inc(p, se.nr_wakeups_local);
2312 else 2340 else
2313 schedstat_inc(p, se.nr_wakeups_remote); 2341 schedstat_inc(p, se.nr_wakeups_remote);
2314 update_rq_clock(rq);
2315 activate_task(rq, p, 1); 2342 activate_task(rq, p, 1);
2316 success = 1; 2343 success = 1;
2317 2344
2318out_running: 2345out_running:
2319 trace_sched_wakeup(rq, p); 2346 trace_sched_wakeup(rq, p, success);
2320 check_preempt_curr(rq, p, sync); 2347 check_preempt_curr(rq, p, sync);
2321 2348
2322 p->state = TASK_RUNNING; 2349 p->state = TASK_RUNNING;
@@ -2449,7 +2476,7 @@ void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2449 p->sched_class->task_new(rq, p); 2476 p->sched_class->task_new(rq, p);
2450 inc_nr_running(rq); 2477 inc_nr_running(rq);
2451 } 2478 }
2452 trace_sched_wakeup_new(rq, p); 2479 trace_sched_wakeup_new(rq, p, 1);
2453 check_preempt_curr(rq, p, 0); 2480 check_preempt_curr(rq, p, 0);
2454#ifdef CONFIG_SMP 2481#ifdef CONFIG_SMP
2455 if (p->sched_class->task_wake_up) 2482 if (p->sched_class->task_wake_up)
@@ -2812,40 +2839,6 @@ static void double_rq_unlock(struct rq *rq1, struct rq *rq2)
2812} 2839}
2813 2840
2814/* 2841/*
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. 2842 * 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 2843 * This is accomplished by forcing the cpu_allowed mask to only
2851 * allow dest_cpu, which will force the cpu onto dest_cpu. Then 2844 * allow dest_cpu, which will force the cpu onto dest_cpu. Then
@@ -2858,11 +2851,10 @@ static void sched_migrate_task(struct task_struct *p, int dest_cpu)
2858 struct rq *rq; 2851 struct rq *rq;
2859 2852
2860 rq = task_rq_lock(p, &flags); 2853 rq = task_rq_lock(p, &flags);
2861 if (!cpu_isset(dest_cpu, p->cpus_allowed) 2854 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)
2862 || unlikely(!cpu_active(dest_cpu))) 2855 || unlikely(!cpu_active(dest_cpu)))
2863 goto out; 2856 goto out;
2864 2857
2865 trace_sched_migrate_task(rq, p, dest_cpu);
2866 /* force the process onto the specified CPU */ 2858 /* force the process onto the specified CPU */
2867 if (migrate_task(p, dest_cpu, &req)) { 2859 if (migrate_task(p, dest_cpu, &req)) {
2868 /* Need to wait for migration thread (might exit: take ref). */ 2860 /* Need to wait for migration thread (might exit: take ref). */
@@ -2924,7 +2916,7 @@ 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 2916 * 2) cannot be migrated to this CPU due to cpus_allowed, or
2925 * 3) are cache-hot on their current CPU. 2917 * 3) are cache-hot on their current CPU.
2926 */ 2918 */
2927 if (!cpu_isset(this_cpu, p->cpus_allowed)) { 2919 if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) {
2928 schedstat_inc(p, se.nr_failed_migrations_affine); 2920 schedstat_inc(p, se.nr_failed_migrations_affine);
2929 return 0; 2921 return 0;
2930 } 2922 }
@@ -3099,7 +3091,7 @@ static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
3099static struct sched_group * 3091static struct sched_group *
3100find_busiest_group(struct sched_domain *sd, int this_cpu, 3092find_busiest_group(struct sched_domain *sd, int this_cpu,
3101 unsigned long *imbalance, enum cpu_idle_type idle, 3093 unsigned long *imbalance, enum cpu_idle_type idle,
3102 int *sd_idle, const cpumask_t *cpus, int *balance) 3094 int *sd_idle, const struct cpumask *cpus, int *balance)
3103{ 3095{
3104 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; 3096 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
3105 unsigned long max_load, avg_load, total_load, this_load, total_pwr; 3097 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
@@ -3135,10 +3127,11 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3135 unsigned long sum_avg_load_per_task; 3127 unsigned long sum_avg_load_per_task;
3136 unsigned long avg_load_per_task; 3128 unsigned long avg_load_per_task;
3137 3129
3138 local_group = cpu_isset(this_cpu, group->cpumask); 3130 local_group = cpumask_test_cpu(this_cpu,
3131 sched_group_cpus(group));
3139 3132
3140 if (local_group) 3133 if (local_group)
3141 balance_cpu = first_cpu(group->cpumask); 3134 balance_cpu = cpumask_first(sched_group_cpus(group));
3142 3135
3143 /* Tally up the load of all CPUs in the group */ 3136 /* Tally up the load of all CPUs in the group */
3144 sum_weighted_load = sum_nr_running = avg_load = 0; 3137 sum_weighted_load = sum_nr_running = avg_load = 0;
@@ -3147,13 +3140,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3147 max_cpu_load = 0; 3140 max_cpu_load = 0;
3148 min_cpu_load = ~0UL; 3141 min_cpu_load = ~0UL;
3149 3142
3150 for_each_cpu_mask_nr(i, group->cpumask) { 3143 for_each_cpu_and(i, sched_group_cpus(group), cpus) {
3151 struct rq *rq; 3144 struct rq *rq = cpu_rq(i);
3152
3153 if (!cpu_isset(i, *cpus))
3154 continue;
3155
3156 rq = cpu_rq(i);
3157 3145
3158 if (*sd_idle && rq->nr_running) 3146 if (*sd_idle && rq->nr_running)
3159 *sd_idle = 0; 3147 *sd_idle = 0;
@@ -3264,8 +3252,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3264 */ 3252 */
3265 if ((sum_nr_running < min_nr_running) || 3253 if ((sum_nr_running < min_nr_running) ||
3266 (sum_nr_running == min_nr_running && 3254 (sum_nr_running == min_nr_running &&
3267 first_cpu(group->cpumask) < 3255 cpumask_first(sched_group_cpus(group)) >
3268 first_cpu(group_min->cpumask))) { 3256 cpumask_first(sched_group_cpus(group_min)))) {
3269 group_min = group; 3257 group_min = group;
3270 min_nr_running = sum_nr_running; 3258 min_nr_running = sum_nr_running;
3271 min_load_per_task = sum_weighted_load / 3259 min_load_per_task = sum_weighted_load /
@@ -3280,8 +3268,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
3280 if (sum_nr_running <= group_capacity - 1) { 3268 if (sum_nr_running <= group_capacity - 1) {
3281 if (sum_nr_running > leader_nr_running || 3269 if (sum_nr_running > leader_nr_running ||
3282 (sum_nr_running == leader_nr_running && 3270 (sum_nr_running == leader_nr_running &&
3283 first_cpu(group->cpumask) > 3271 cpumask_first(sched_group_cpus(group)) <
3284 first_cpu(group_leader->cpumask))) { 3272 cpumask_first(sched_group_cpus(group_leader)))) {
3285 group_leader = group; 3273 group_leader = group;
3286 leader_nr_running = sum_nr_running; 3274 leader_nr_running = sum_nr_running;
3287 } 3275 }
@@ -3407,6 +3395,10 @@ out_balanced:
3407 3395
3408 if (this == group_leader && group_leader != group_min) { 3396 if (this == group_leader && group_leader != group_min) {
3409 *imbalance = min_load_per_task; 3397 *imbalance = min_load_per_task;
3398 if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) {
3399 cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu =
3400 cpumask_first(sched_group_cpus(group_leader));
3401 }
3410 return group_min; 3402 return group_min;
3411 } 3403 }
3412#endif 3404#endif
@@ -3420,16 +3412,16 @@ ret:
3420 */ 3412 */
3421static struct rq * 3413static struct rq *
3422find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, 3414find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
3423 unsigned long imbalance, const cpumask_t *cpus) 3415 unsigned long imbalance, const struct cpumask *cpus)
3424{ 3416{
3425 struct rq *busiest = NULL, *rq; 3417 struct rq *busiest = NULL, *rq;
3426 unsigned long max_load = 0; 3418 unsigned long max_load = 0;
3427 int i; 3419 int i;
3428 3420
3429 for_each_cpu_mask_nr(i, group->cpumask) { 3421 for_each_cpu(i, sched_group_cpus(group)) {
3430 unsigned long wl; 3422 unsigned long wl;
3431 3423
3432 if (!cpu_isset(i, *cpus)) 3424 if (!cpumask_test_cpu(i, cpus))
3433 continue; 3425 continue;
3434 3426
3435 rq = cpu_rq(i); 3427 rq = cpu_rq(i);
@@ -3459,7 +3451,7 @@ find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
3459 */ 3451 */
3460static int load_balance(int this_cpu, struct rq *this_rq, 3452static int load_balance(int this_cpu, struct rq *this_rq,
3461 struct sched_domain *sd, enum cpu_idle_type idle, 3453 struct sched_domain *sd, enum cpu_idle_type idle,
3462 int *balance, cpumask_t *cpus) 3454 int *balance, struct cpumask *cpus)
3463{ 3455{
3464 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; 3456 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
3465 struct sched_group *group; 3457 struct sched_group *group;
@@ -3467,7 +3459,7 @@ static int load_balance(int this_cpu, struct rq *this_rq,
3467 struct rq *busiest; 3459 struct rq *busiest;
3468 unsigned long flags; 3460 unsigned long flags;
3469 3461
3470 cpus_setall(*cpus); 3462 cpumask_setall(cpus);
3471 3463
3472 /* 3464 /*
3473 * When power savings policy is enabled for the parent domain, idle 3465 * When power savings policy is enabled for the parent domain, idle
@@ -3527,8 +3519,8 @@ redo:
3527 3519
3528 /* All tasks on this runqueue were pinned by CPU affinity */ 3520 /* All tasks on this runqueue were pinned by CPU affinity */
3529 if (unlikely(all_pinned)) { 3521 if (unlikely(all_pinned)) {
3530 cpu_clear(cpu_of(busiest), *cpus); 3522 cpumask_clear_cpu(cpu_of(busiest), cpus);
3531 if (!cpus_empty(*cpus)) 3523 if (!cpumask_empty(cpus))
3532 goto redo; 3524 goto redo;
3533 goto out_balanced; 3525 goto out_balanced;
3534 } 3526 }
@@ -3545,7 +3537,8 @@ redo:
3545 /* don't kick the migration_thread, if the curr 3537 /* don't kick the migration_thread, if the curr
3546 * task on busiest cpu can't be moved to this_cpu 3538 * task on busiest cpu can't be moved to this_cpu
3547 */ 3539 */
3548 if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { 3540 if (!cpumask_test_cpu(this_cpu,
3541 &busiest->curr->cpus_allowed)) {
3549 spin_unlock_irqrestore(&busiest->lock, flags); 3542 spin_unlock_irqrestore(&busiest->lock, flags);
3550 all_pinned = 1; 3543 all_pinned = 1;
3551 goto out_one_pinned; 3544 goto out_one_pinned;
@@ -3620,7 +3613,7 @@ out:
3620 */ 3613 */
3621static int 3614static int
3622load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, 3615load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd,
3623 cpumask_t *cpus) 3616 struct cpumask *cpus)
3624{ 3617{
3625 struct sched_group *group; 3618 struct sched_group *group;
3626 struct rq *busiest = NULL; 3619 struct rq *busiest = NULL;
@@ -3629,7 +3622,7 @@ load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd,
3629 int sd_idle = 0; 3622 int sd_idle = 0;
3630 int all_pinned = 0; 3623 int all_pinned = 0;
3631 3624
3632 cpus_setall(*cpus); 3625 cpumask_setall(cpus);
3633 3626
3634 /* 3627 /*
3635 * When power savings policy is enabled for the parent domain, idle 3628 * When power savings policy is enabled for the parent domain, idle
@@ -3673,17 +3666,71 @@ redo:
3673 double_unlock_balance(this_rq, busiest); 3666 double_unlock_balance(this_rq, busiest);
3674 3667
3675 if (unlikely(all_pinned)) { 3668 if (unlikely(all_pinned)) {
3676 cpu_clear(cpu_of(busiest), *cpus); 3669 cpumask_clear_cpu(cpu_of(busiest), cpus);
3677 if (!cpus_empty(*cpus)) 3670 if (!cpumask_empty(cpus))
3678 goto redo; 3671 goto redo;
3679 } 3672 }
3680 } 3673 }
3681 3674
3682 if (!ld_moved) { 3675 if (!ld_moved) {
3676 int active_balance = 0;
3677
3683 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); 3678 schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]);
3684 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3679 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3685 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3680 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3686 return -1; 3681 return -1;
3682
3683 if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP)
3684 return -1;
3685
3686 if (sd->nr_balance_failed++ < 2)
3687 return -1;
3688
3689 /*
3690 * The only task running in a non-idle cpu can be moved to this
3691 * cpu in an attempt to completely freeup the other CPU
3692 * package. The same method used to move task in load_balance()
3693 * have been extended for load_balance_newidle() to speedup
3694 * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2)
3695 *
3696 * The package power saving logic comes from
3697 * find_busiest_group(). If there are no imbalance, then
3698 * f_b_g() will return NULL. However when sched_mc={1,2} then
3699 * f_b_g() will select a group from which a running task may be
3700 * pulled to this cpu in order to make the other package idle.
3701 * If there is no opportunity to make a package idle and if
3702 * there are no imbalance, then f_b_g() will return NULL and no
3703 * action will be taken in load_balance_newidle().
3704 *
3705 * Under normal task pull operation due to imbalance, there
3706 * will be more than one task in the source run queue and
3707 * move_tasks() will succeed. ld_moved will be true and this
3708 * active balance code will not be triggered.
3709 */
3710
3711 /* Lock busiest in correct order while this_rq is held */
3712 double_lock_balance(this_rq, busiest);
3713
3714 /*
3715 * don't kick the migration_thread, if the curr
3716 * task on busiest cpu can't be moved to this_cpu
3717 */
3718 if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
3719 double_unlock_balance(this_rq, busiest);
3720 all_pinned = 1;
3721 return ld_moved;
3722 }
3723
3724 if (!busiest->active_balance) {
3725 busiest->active_balance = 1;
3726 busiest->push_cpu = this_cpu;
3727 active_balance = 1;
3728 }
3729
3730 double_unlock_balance(this_rq, busiest);
3731 if (active_balance)
3732 wake_up_process(busiest->migration_thread);
3733
3687 } else 3734 } else
3688 sd->nr_balance_failed = 0; 3735 sd->nr_balance_failed = 0;
3689 3736
@@ -3707,9 +3754,12 @@ out_balanced:
3707static void idle_balance(int this_cpu, struct rq *this_rq) 3754static void idle_balance(int this_cpu, struct rq *this_rq)
3708{ 3755{
3709 struct sched_domain *sd; 3756 struct sched_domain *sd;
3710 int pulled_task = -1; 3757 int pulled_task = 0;
3711 unsigned long next_balance = jiffies + HZ; 3758 unsigned long next_balance = jiffies + HZ;
3712 cpumask_t tmpmask; 3759 cpumask_var_t tmpmask;
3760
3761 if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC))
3762 return;
3713 3763
3714 for_each_domain(this_cpu, sd) { 3764 for_each_domain(this_cpu, sd) {
3715 unsigned long interval; 3765 unsigned long interval;
@@ -3720,7 +3770,7 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3720 if (sd->flags & SD_BALANCE_NEWIDLE) 3770 if (sd->flags & SD_BALANCE_NEWIDLE)
3721 /* If we've pulled tasks over stop searching: */ 3771 /* If we've pulled tasks over stop searching: */
3722 pulled_task = load_balance_newidle(this_cpu, this_rq, 3772 pulled_task = load_balance_newidle(this_cpu, this_rq,
3723 sd, &tmpmask); 3773 sd, tmpmask);
3724 3774
3725 interval = msecs_to_jiffies(sd->balance_interval); 3775 interval = msecs_to_jiffies(sd->balance_interval);
3726 if (time_after(next_balance, sd->last_balance + interval)) 3776 if (time_after(next_balance, sd->last_balance + interval))
@@ -3735,6 +3785,7 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3735 */ 3785 */
3736 this_rq->next_balance = next_balance; 3786 this_rq->next_balance = next_balance;
3737 } 3787 }
3788 free_cpumask_var(tmpmask);
3738} 3789}
3739 3790
3740/* 3791/*
@@ -3772,7 +3823,7 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
3772 /* Search for an sd spanning us and the target CPU. */ 3823 /* Search for an sd spanning us and the target CPU. */
3773 for_each_domain(target_cpu, sd) { 3824 for_each_domain(target_cpu, sd) {
3774 if ((sd->flags & SD_LOAD_BALANCE) && 3825 if ((sd->flags & SD_LOAD_BALANCE) &&
3775 cpu_isset(busiest_cpu, sd->span)) 3826 cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
3776 break; 3827 break;
3777 } 3828 }
3778 3829
@@ -3791,10 +3842,9 @@ static void active_load_balance(struct rq *busiest_rq, int busiest_cpu)
3791#ifdef CONFIG_NO_HZ 3842#ifdef CONFIG_NO_HZ
3792static struct { 3843static struct {
3793 atomic_t load_balancer; 3844 atomic_t load_balancer;
3794 cpumask_t cpu_mask; 3845 cpumask_var_t cpu_mask;
3795} nohz ____cacheline_aligned = { 3846} nohz ____cacheline_aligned = {
3796 .load_balancer = ATOMIC_INIT(-1), 3847 .load_balancer = ATOMIC_INIT(-1),
3797 .cpu_mask = CPU_MASK_NONE,
3798}; 3848};
3799 3849
3800/* 3850/*
@@ -3822,7 +3872,7 @@ int select_nohz_load_balancer(int stop_tick)
3822 int cpu = smp_processor_id(); 3872 int cpu = smp_processor_id();
3823 3873
3824 if (stop_tick) { 3874 if (stop_tick) {
3825 cpu_set(cpu, nohz.cpu_mask); 3875 cpumask_set_cpu(cpu, nohz.cpu_mask);
3826 cpu_rq(cpu)->in_nohz_recently = 1; 3876 cpu_rq(cpu)->in_nohz_recently = 1;
3827 3877
3828 /* 3878 /*
@@ -3836,7 +3886,7 @@ int select_nohz_load_balancer(int stop_tick)
3836 } 3886 }
3837 3887
3838 /* time for ilb owner also to sleep */ 3888 /* time for ilb owner also to sleep */
3839 if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { 3889 if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) {
3840 if (atomic_read(&nohz.load_balancer) == cpu) 3890 if (atomic_read(&nohz.load_balancer) == cpu)
3841 atomic_set(&nohz.load_balancer, -1); 3891 atomic_set(&nohz.load_balancer, -1);
3842 return 0; 3892 return 0;
@@ -3849,10 +3899,10 @@ int select_nohz_load_balancer(int stop_tick)
3849 } else if (atomic_read(&nohz.load_balancer) == cpu) 3899 } else if (atomic_read(&nohz.load_balancer) == cpu)
3850 return 1; 3900 return 1;
3851 } else { 3901 } else {
3852 if (!cpu_isset(cpu, nohz.cpu_mask)) 3902 if (!cpumask_test_cpu(cpu, nohz.cpu_mask))
3853 return 0; 3903 return 0;
3854 3904
3855 cpu_clear(cpu, nohz.cpu_mask); 3905 cpumask_clear_cpu(cpu, nohz.cpu_mask);
3856 3906
3857 if (atomic_read(&nohz.load_balancer) == cpu) 3907 if (atomic_read(&nohz.load_balancer) == cpu)
3858 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) 3908 if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu)
@@ -3880,7 +3930,11 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
3880 unsigned long next_balance = jiffies + 60*HZ; 3930 unsigned long next_balance = jiffies + 60*HZ;
3881 int update_next_balance = 0; 3931 int update_next_balance = 0;
3882 int need_serialize; 3932 int need_serialize;
3883 cpumask_t tmp; 3933 cpumask_var_t tmp;
3934
3935 /* Fails alloc? Rebalancing probably not a priority right now. */
3936 if (!alloc_cpumask_var(&tmp, GFP_ATOMIC))
3937 return;
3884 3938
3885 for_each_domain(cpu, sd) { 3939 for_each_domain(cpu, sd) {
3886 if (!(sd->flags & SD_LOAD_BALANCE)) 3940 if (!(sd->flags & SD_LOAD_BALANCE))
@@ -3905,7 +3959,7 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
3905 } 3959 }
3906 3960
3907 if (time_after_eq(jiffies, sd->last_balance + interval)) { 3961 if (time_after_eq(jiffies, sd->last_balance + interval)) {
3908 if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { 3962 if (load_balance(cpu, rq, sd, idle, &balance, tmp)) {
3909 /* 3963 /*
3910 * We've pulled tasks over so either we're no 3964 * We've pulled tasks over so either we're no
3911 * longer idle, or one of our SMT siblings is 3965 * longer idle, or one of our SMT siblings is
@@ -3939,6 +3993,8 @@ out:
3939 */ 3993 */
3940 if (likely(update_next_balance)) 3994 if (likely(update_next_balance))
3941 rq->next_balance = next_balance; 3995 rq->next_balance = next_balance;
3996
3997 free_cpumask_var(tmp);
3942} 3998}
3943 3999
3944/* 4000/*
@@ -3963,12 +4019,13 @@ static void run_rebalance_domains(struct softirq_action *h)
3963 */ 4019 */
3964 if (this_rq->idle_at_tick && 4020 if (this_rq->idle_at_tick &&
3965 atomic_read(&nohz.load_balancer) == this_cpu) { 4021 atomic_read(&nohz.load_balancer) == this_cpu) {
3966 cpumask_t cpus = nohz.cpu_mask;
3967 struct rq *rq; 4022 struct rq *rq;
3968 int balance_cpu; 4023 int balance_cpu;
3969 4024
3970 cpu_clear(this_cpu, cpus); 4025 for_each_cpu(balance_cpu, nohz.cpu_mask) {
3971 for_each_cpu_mask_nr(balance_cpu, cpus) { 4026 if (balance_cpu == this_cpu)
4027 continue;
4028
3972 /* 4029 /*
3973 * If this cpu gets work to do, stop the load balancing 4030 * If this cpu gets work to do, stop the load balancing
3974 * work being done for other cpus. Next load 4031 * work being done for other cpus. Next load
@@ -4006,7 +4063,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4006 rq->in_nohz_recently = 0; 4063 rq->in_nohz_recently = 0;
4007 4064
4008 if (atomic_read(&nohz.load_balancer) == cpu) { 4065 if (atomic_read(&nohz.load_balancer) == cpu) {
4009 cpu_clear(cpu, nohz.cpu_mask); 4066 cpumask_clear_cpu(cpu, nohz.cpu_mask);
4010 atomic_set(&nohz.load_balancer, -1); 4067 atomic_set(&nohz.load_balancer, -1);
4011 } 4068 }
4012 4069
@@ -4019,7 +4076,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4019 * TBD: Traverse the sched domains and nominate 4076 * TBD: Traverse the sched domains and nominate
4020 * the nearest cpu in the nohz.cpu_mask. 4077 * the nearest cpu in the nohz.cpu_mask.
4021 */ 4078 */
4022 int ilb = first_cpu(nohz.cpu_mask); 4079 int ilb = cpumask_first(nohz.cpu_mask);
4023 4080
4024 if (ilb < nr_cpu_ids) 4081 if (ilb < nr_cpu_ids)
4025 resched_cpu(ilb); 4082 resched_cpu(ilb);
@@ -4031,7 +4088,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4031 * cpus with ticks stopped, is it time for that to stop? 4088 * cpus with ticks stopped, is it time for that to stop?
4032 */ 4089 */
4033 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && 4090 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu &&
4034 cpus_weight(nohz.cpu_mask) == num_online_cpus()) { 4091 cpumask_weight(nohz.cpu_mask) == num_online_cpus()) {
4035 resched_cpu(cpu); 4092 resched_cpu(cpu);
4036 return; 4093 return;
4037 } 4094 }
@@ -4041,7 +4098,7 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4041 * someone else, then no need raise the SCHED_SOFTIRQ 4098 * someone else, then no need raise the SCHED_SOFTIRQ
4042 */ 4099 */
4043 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && 4100 if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu &&
4044 cpu_isset(cpu, nohz.cpu_mask)) 4101 cpumask_test_cpu(cpu, nohz.cpu_mask))
4045 return; 4102 return;
4046#endif 4103#endif
4047 if (time_after_eq(jiffies, rq->next_balance)) 4104 if (time_after_eq(jiffies, rq->next_balance))
@@ -4093,13 +4150,17 @@ unsigned long long task_delta_exec(struct task_struct *p)
4093 * Account user cpu time to a process. 4150 * Account user cpu time to a process.
4094 * @p: the process that the cpu time gets accounted to 4151 * @p: the process that the cpu time gets accounted to
4095 * @cputime: the cpu time spent in user space since the last update 4152 * @cputime: the cpu time spent in user space since the last update
4153 * @cputime_scaled: cputime scaled by cpu frequency
4096 */ 4154 */
4097void account_user_time(struct task_struct *p, cputime_t cputime) 4155void account_user_time(struct task_struct *p, cputime_t cputime,
4156 cputime_t cputime_scaled)
4098{ 4157{
4099 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4158 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4100 cputime64_t tmp; 4159 cputime64_t tmp;
4101 4160
4161 /* Add user time to process. */
4102 p->utime = cputime_add(p->utime, cputime); 4162 p->utime = cputime_add(p->utime, cputime);
4163 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
4103 account_group_user_time(p, cputime); 4164 account_group_user_time(p, cputime);
4104 4165
4105 /* Add user time to cpustat. */ 4166 /* Add user time to cpustat. */
@@ -4116,51 +4177,48 @@ void account_user_time(struct task_struct *p, cputime_t cputime)
4116 * Account guest cpu time to a process. 4177 * Account guest cpu time to a process.
4117 * @p: the process that the cpu time gets accounted to 4178 * @p: the process that the cpu time gets accounted to
4118 * @cputime: the cpu time spent in virtual machine since the last update 4179 * @cputime: the cpu time spent in virtual machine since the last update
4180 * @cputime_scaled: cputime scaled by cpu frequency
4119 */ 4181 */
4120static void account_guest_time(struct task_struct *p, cputime_t cputime) 4182static void account_guest_time(struct task_struct *p, cputime_t cputime,
4183 cputime_t cputime_scaled)
4121{ 4184{
4122 cputime64_t tmp; 4185 cputime64_t tmp;
4123 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4186 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4124 4187
4125 tmp = cputime_to_cputime64(cputime); 4188 tmp = cputime_to_cputime64(cputime);
4126 4189
4190 /* Add guest time to process. */
4127 p->utime = cputime_add(p->utime, cputime); 4191 p->utime = cputime_add(p->utime, cputime);
4192 p->utimescaled = cputime_add(p->utimescaled, cputime_scaled);
4128 account_group_user_time(p, cputime); 4193 account_group_user_time(p, cputime);
4129 p->gtime = cputime_add(p->gtime, cputime); 4194 p->gtime = cputime_add(p->gtime, cputime);
4130 4195
4196 /* Add guest time to cpustat. */
4131 cpustat->user = cputime64_add(cpustat->user, tmp); 4197 cpustat->user = cputime64_add(cpustat->user, tmp);
4132 cpustat->guest = cputime64_add(cpustat->guest, tmp); 4198 cpustat->guest = cputime64_add(cpustat->guest, tmp);
4133} 4199}
4134 4200
4135/* 4201/*
4136 * Account scaled user cpu time to a process.
4137 * @p: the process that the cpu time gets accounted to
4138 * @cputime: the cpu time spent in user space since the last update
4139 */
4140void account_user_time_scaled(struct task_struct *p, cputime_t cputime)
4141{
4142 p->utimescaled = cputime_add(p->utimescaled, cputime);
4143}
4144
4145/*
4146 * Account system cpu time to a process. 4202 * Account system cpu time to a process.
4147 * @p: the process that the cpu time gets accounted to 4203 * @p: the process that the cpu time gets accounted to
4148 * @hardirq_offset: the offset to subtract from hardirq_count() 4204 * @hardirq_offset: the offset to subtract from hardirq_count()
4149 * @cputime: the cpu time spent in kernel space since the last update 4205 * @cputime: the cpu time spent in kernel space since the last update
4206 * @cputime_scaled: cputime scaled by cpu frequency
4150 */ 4207 */
4151void account_system_time(struct task_struct *p, int hardirq_offset, 4208void account_system_time(struct task_struct *p, int hardirq_offset,
4152 cputime_t cputime) 4209 cputime_t cputime, cputime_t cputime_scaled)
4153{ 4210{
4154 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4211 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4155 struct rq *rq = this_rq();
4156 cputime64_t tmp; 4212 cputime64_t tmp;
4157 4213
4158 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { 4214 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
4159 account_guest_time(p, cputime); 4215 account_guest_time(p, cputime, cputime_scaled);
4160 return; 4216 return;
4161 } 4217 }
4162 4218
4219 /* Add system time to process. */
4163 p->stime = cputime_add(p->stime, cputime); 4220 p->stime = cputime_add(p->stime, cputime);
4221 p->stimescaled = cputime_add(p->stimescaled, cputime_scaled);
4164 account_group_system_time(p, cputime); 4222 account_group_system_time(p, cputime);
4165 4223
4166 /* Add system time to cpustat. */ 4224 /* Add system time to cpustat. */
@@ -4169,50 +4227,85 @@ void account_system_time(struct task_struct *p, int hardirq_offset,
4169 cpustat->irq = cputime64_add(cpustat->irq, tmp); 4227 cpustat->irq = cputime64_add(cpustat->irq, tmp);
4170 else if (softirq_count()) 4228 else if (softirq_count())
4171 cpustat->softirq = cputime64_add(cpustat->softirq, tmp); 4229 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
4172 else if (p != rq->idle)
4173 cpustat->system = cputime64_add(cpustat->system, tmp);
4174 else if (atomic_read(&rq->nr_iowait) > 0)
4175 cpustat->iowait = cputime64_add(cpustat->iowait, tmp);
4176 else 4230 else
4177 cpustat->idle = cputime64_add(cpustat->idle, tmp); 4231 cpustat->system = cputime64_add(cpustat->system, tmp);
4232
4178 /* Account for system time used */ 4233 /* Account for system time used */
4179 acct_update_integrals(p); 4234 acct_update_integrals(p);
4180} 4235}
4181 4236
4182/* 4237/*
4183 * Account scaled system cpu time to a process. 4238 * Account for involuntary wait time.
4184 * @p: the process that the cpu time gets accounted to 4239 * @steal: the cpu time spent in involuntary wait
4185 * @hardirq_offset: the offset to subtract from hardirq_count()
4186 * @cputime: the cpu time spent in kernel space since the last update
4187 */ 4240 */
4188void account_system_time_scaled(struct task_struct *p, cputime_t cputime) 4241void account_steal_time(cputime_t cputime)
4189{ 4242{
4190 p->stimescaled = cputime_add(p->stimescaled, cputime); 4243 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4244 cputime64_t cputime64 = cputime_to_cputime64(cputime);
4245
4246 cpustat->steal = cputime64_add(cpustat->steal, cputime64);
4191} 4247}
4192 4248
4193/* 4249/*
4194 * Account for involuntary wait time. 4250 * Account for idle time.
4195 * @p: the process from which the cpu time has been stolen 4251 * @cputime: the cpu time spent in idle wait
4196 * @steal: the cpu time spent in involuntary wait
4197 */ 4252 */
4198void account_steal_time(struct task_struct *p, cputime_t steal) 4253void account_idle_time(cputime_t cputime)
4199{ 4254{
4200 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; 4255 struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat;
4201 cputime64_t tmp = cputime_to_cputime64(steal); 4256 cputime64_t cputime64 = cputime_to_cputime64(cputime);
4202 struct rq *rq = this_rq(); 4257 struct rq *rq = this_rq();
4203 4258
4204 if (p == rq->idle) { 4259 if (atomic_read(&rq->nr_iowait) > 0)
4205 p->stime = cputime_add(p->stime, steal); 4260 cpustat->iowait = cputime64_add(cpustat->iowait, cputime64);
4206 account_group_system_time(p, steal); 4261 else
4207 if (atomic_read(&rq->nr_iowait) > 0) 4262 cpustat->idle = cputime64_add(cpustat->idle, cputime64);
4208 cpustat->iowait = cputime64_add(cpustat->iowait, tmp); 4263}
4209 else 4264
4210 cpustat->idle = cputime64_add(cpustat->idle, tmp); 4265#ifndef CONFIG_VIRT_CPU_ACCOUNTING
4211 } else 4266
4212 cpustat->steal = cputime64_add(cpustat->steal, tmp); 4267/*
4268 * Account a single tick of cpu time.
4269 * @p: the process that the cpu time gets accounted to
4270 * @user_tick: indicates if the tick is a user or a system tick
4271 */
4272void account_process_tick(struct task_struct *p, int user_tick)
4273{
4274 cputime_t one_jiffy = jiffies_to_cputime(1);
4275 cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy);
4276 struct rq *rq = this_rq();
4277
4278 if (user_tick)
4279 account_user_time(p, one_jiffy, one_jiffy_scaled);
4280 else if (p != rq->idle)
4281 account_system_time(p, HARDIRQ_OFFSET, one_jiffy,
4282 one_jiffy_scaled);
4283 else
4284 account_idle_time(one_jiffy);
4285}
4286
4287/*
4288 * Account multiple ticks of steal time.
4289 * @p: the process from which the cpu time has been stolen
4290 * @ticks: number of stolen ticks
4291 */
4292void account_steal_ticks(unsigned long ticks)
4293{
4294 account_steal_time(jiffies_to_cputime(ticks));
4213} 4295}
4214 4296
4215/* 4297/*
4298 * Account multiple ticks of idle time.
4299 * @ticks: number of stolen ticks
4300 */
4301void account_idle_ticks(unsigned long ticks)
4302{
4303 account_idle_time(jiffies_to_cputime(ticks));
4304}
4305
4306#endif
4307
4308/*
4216 * Use precise platform statistics if available: 4309 * Use precise platform statistics if available:
4217 */ 4310 */
4218#ifdef CONFIG_VIRT_CPU_ACCOUNTING 4311#ifdef CONFIG_VIRT_CPU_ACCOUNTING
@@ -4339,7 +4432,7 @@ void __kprobes sub_preempt_count(int val)
4339 /* 4432 /*
4340 * Underflow? 4433 * Underflow?
4341 */ 4434 */
4342 if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) 4435 if (DEBUG_LOCKS_WARN_ON(val > preempt_count() - (!!kernel_locked())))
4343 return; 4436 return;
4344 /* 4437 /*
4345 * Is the spinlock portion underflowing? 4438 * Is the spinlock portion underflowing?
@@ -5134,6 +5227,22 @@ __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
5134 set_load_weight(p); 5227 set_load_weight(p);
5135} 5228}
5136 5229
5230/*
5231 * check the target process has a UID that matches the current process's
5232 */
5233static bool check_same_owner(struct task_struct *p)
5234{
5235 const struct cred *cred = current_cred(), *pcred;
5236 bool match;
5237
5238 rcu_read_lock();
5239 pcred = __task_cred(p);
5240 match = (cred->euid == pcred->euid ||
5241 cred->euid == pcred->uid);
5242 rcu_read_unlock();
5243 return match;
5244}
5245
5137static int __sched_setscheduler(struct task_struct *p, int policy, 5246static int __sched_setscheduler(struct task_struct *p, int policy,
5138 struct sched_param *param, bool user) 5247 struct sched_param *param, bool user)
5139{ 5248{
@@ -5193,8 +5302,7 @@ recheck:
5193 return -EPERM; 5302 return -EPERM;
5194 5303
5195 /* can't change other user's priorities */ 5304 /* can't change other user's priorities */
5196 if ((current->euid != p->euid) && 5305 if (!check_same_owner(p))
5197 (current->euid != p->uid))
5198 return -EPERM; 5306 return -EPERM;
5199 } 5307 }
5200 5308
@@ -5400,10 +5508,9 @@ out_unlock:
5400 return retval; 5508 return retval;
5401} 5509}
5402 5510
5403long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) 5511long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
5404{ 5512{
5405 cpumask_t cpus_allowed; 5513 cpumask_var_t cpus_allowed, new_mask;
5406 cpumask_t new_mask = *in_mask;
5407 struct task_struct *p; 5514 struct task_struct *p;
5408 int retval; 5515 int retval;
5409 5516
@@ -5425,46 +5532,57 @@ long sched_setaffinity(pid_t pid, const cpumask_t *in_mask)
5425 get_task_struct(p); 5532 get_task_struct(p);
5426 read_unlock(&tasklist_lock); 5533 read_unlock(&tasklist_lock);
5427 5534
5535 if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
5536 retval = -ENOMEM;
5537 goto out_put_task;
5538 }
5539 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
5540 retval = -ENOMEM;
5541 goto out_free_cpus_allowed;
5542 }
5428 retval = -EPERM; 5543 retval = -EPERM;
5429 if ((current->euid != p->euid) && (current->euid != p->uid) && 5544 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
5430 !capable(CAP_SYS_NICE))
5431 goto out_unlock; 5545 goto out_unlock;
5432 5546
5433 retval = security_task_setscheduler(p, 0, NULL); 5547 retval = security_task_setscheduler(p, 0, NULL);
5434 if (retval) 5548 if (retval)
5435 goto out_unlock; 5549 goto out_unlock;
5436 5550
5437 cpuset_cpus_allowed(p, &cpus_allowed); 5551 cpuset_cpus_allowed(p, cpus_allowed);
5438 cpus_and(new_mask, new_mask, cpus_allowed); 5552 cpumask_and(new_mask, in_mask, cpus_allowed);
5439 again: 5553 again:
5440 retval = set_cpus_allowed_ptr(p, &new_mask); 5554 retval = set_cpus_allowed_ptr(p, new_mask);
5441 5555
5442 if (!retval) { 5556 if (!retval) {
5443 cpuset_cpus_allowed(p, &cpus_allowed); 5557 cpuset_cpus_allowed(p, cpus_allowed);
5444 if (!cpus_subset(new_mask, cpus_allowed)) { 5558 if (!cpumask_subset(new_mask, cpus_allowed)) {
5445 /* 5559 /*
5446 * We must have raced with a concurrent cpuset 5560 * We must have raced with a concurrent cpuset
5447 * update. Just reset the cpus_allowed to the 5561 * update. Just reset the cpus_allowed to the
5448 * cpuset's cpus_allowed 5562 * cpuset's cpus_allowed
5449 */ 5563 */
5450 new_mask = cpus_allowed; 5564 cpumask_copy(new_mask, cpus_allowed);
5451 goto again; 5565 goto again;
5452 } 5566 }
5453 } 5567 }
5454out_unlock: 5568out_unlock:
5569 free_cpumask_var(new_mask);
5570out_free_cpus_allowed:
5571 free_cpumask_var(cpus_allowed);
5572out_put_task:
5455 put_task_struct(p); 5573 put_task_struct(p);
5456 put_online_cpus(); 5574 put_online_cpus();
5457 return retval; 5575 return retval;
5458} 5576}
5459 5577
5460static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, 5578static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
5461 cpumask_t *new_mask) 5579 struct cpumask *new_mask)
5462{ 5580{
5463 if (len < sizeof(cpumask_t)) { 5581 if (len < cpumask_size())
5464 memset(new_mask, 0, sizeof(cpumask_t)); 5582 cpumask_clear(new_mask);
5465 } else if (len > sizeof(cpumask_t)) { 5583 else if (len > cpumask_size())
5466 len = sizeof(cpumask_t); 5584 len = cpumask_size();
5467 } 5585
5468 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; 5586 return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
5469} 5587}
5470 5588
@@ -5477,17 +5595,20 @@ 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, 5595asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len,
5478 unsigned long __user *user_mask_ptr) 5596 unsigned long __user *user_mask_ptr)
5479{ 5597{
5480 cpumask_t new_mask; 5598 cpumask_var_t new_mask;
5481 int retval; 5599 int retval;
5482 5600
5483 retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); 5601 if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
5484 if (retval) 5602 return -ENOMEM;
5485 return retval;
5486 5603
5487 return sched_setaffinity(pid, &new_mask); 5604 retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
5605 if (retval == 0)
5606 retval = sched_setaffinity(pid, new_mask);
5607 free_cpumask_var(new_mask);
5608 return retval;
5488} 5609}
5489 5610
5490long sched_getaffinity(pid_t pid, cpumask_t *mask) 5611long sched_getaffinity(pid_t pid, struct cpumask *mask)
5491{ 5612{
5492 struct task_struct *p; 5613 struct task_struct *p;
5493 int retval; 5614 int retval;
@@ -5504,7 +5625,7 @@ long sched_getaffinity(pid_t pid, cpumask_t *mask)
5504 if (retval) 5625 if (retval)
5505 goto out_unlock; 5626 goto out_unlock;
5506 5627
5507 cpus_and(*mask, p->cpus_allowed, cpu_online_map); 5628 cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
5508 5629
5509out_unlock: 5630out_unlock:
5510 read_unlock(&tasklist_lock); 5631 read_unlock(&tasklist_lock);
@@ -5523,19 +5644,24 @@ asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len,
5523 unsigned long __user *user_mask_ptr) 5644 unsigned long __user *user_mask_ptr)
5524{ 5645{
5525 int ret; 5646 int ret;
5526 cpumask_t mask; 5647 cpumask_var_t mask;
5527 5648
5528 if (len < sizeof(cpumask_t)) 5649 if (len < cpumask_size())
5529 return -EINVAL; 5650 return -EINVAL;
5530 5651
5531 ret = sched_getaffinity(pid, &mask); 5652 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
5532 if (ret < 0) 5653 return -ENOMEM;
5533 return ret;
5534 5654
5535 if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) 5655 ret = sched_getaffinity(pid, mask);
5536 return -EFAULT; 5656 if (ret == 0) {
5657 if (copy_to_user(user_mask_ptr, mask, cpumask_size()))
5658 ret = -EFAULT;
5659 else
5660 ret = cpumask_size();
5661 }
5662 free_cpumask_var(mask);
5537 5663
5538 return sizeof(cpumask_t); 5664 return ret;
5539} 5665}
5540 5666
5541/** 5667/**
@@ -5877,7 +6003,7 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5877 idle->se.exec_start = sched_clock(); 6003 idle->se.exec_start = sched_clock();
5878 6004
5879 idle->prio = idle->normal_prio = MAX_PRIO; 6005 idle->prio = idle->normal_prio = MAX_PRIO;
5880 idle->cpus_allowed = cpumask_of_cpu(cpu); 6006 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
5881 __set_task_cpu(idle, cpu); 6007 __set_task_cpu(idle, cpu);
5882 6008
5883 rq->curr = rq->idle = idle; 6009 rq->curr = rq->idle = idle;
@@ -5896,6 +6022,7 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5896 * The idle tasks have their own, simple scheduling class: 6022 * The idle tasks have their own, simple scheduling class:
5897 */ 6023 */
5898 idle->sched_class = &idle_sched_class; 6024 idle->sched_class = &idle_sched_class;
6025 ftrace_graph_init_task(idle);
5899} 6026}
5900 6027
5901/* 6028/*
@@ -5903,9 +6030,9 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5903 * indicates which cpus entered this state. This is used 6030 * indicates which cpus entered this state. This is used
5904 * in the rcu update to wait only for active cpus. For system 6031 * 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 6032 * which do not switch off the HZ timer nohz_cpu_mask should
5906 * always be CPU_MASK_NONE. 6033 * always be CPU_BITS_NONE.
5907 */ 6034 */
5908cpumask_t nohz_cpu_mask = CPU_MASK_NONE; 6035cpumask_var_t nohz_cpu_mask;
5909 6036
5910/* 6037/*
5911 * Increase the granularity value when there are more CPUs, 6038 * Increase the granularity value when there are more CPUs,
@@ -5960,7 +6087,7 @@ static inline void sched_init_granularity(void)
5960 * task must not exit() & deallocate itself prematurely. The 6087 * task must not exit() & deallocate itself prematurely. The
5961 * call is not atomic; no spinlocks may be held. 6088 * call is not atomic; no spinlocks may be held.
5962 */ 6089 */
5963int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) 6090int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
5964{ 6091{
5965 struct migration_req req; 6092 struct migration_req req;
5966 unsigned long flags; 6093 unsigned long flags;
@@ -5968,13 +6095,13 @@ int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask)
5968 int ret = 0; 6095 int ret = 0;
5969 6096
5970 rq = task_rq_lock(p, &flags); 6097 rq = task_rq_lock(p, &flags);
5971 if (!cpus_intersects(*new_mask, cpu_online_map)) { 6098 if (!cpumask_intersects(new_mask, cpu_online_mask)) {
5972 ret = -EINVAL; 6099 ret = -EINVAL;
5973 goto out; 6100 goto out;
5974 } 6101 }
5975 6102
5976 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && 6103 if (unlikely((p->flags & PF_THREAD_BOUND) && p != current &&
5977 !cpus_equal(p->cpus_allowed, *new_mask))) { 6104 !cpumask_equal(&p->cpus_allowed, new_mask))) {
5978 ret = -EINVAL; 6105 ret = -EINVAL;
5979 goto out; 6106 goto out;
5980 } 6107 }
@@ -5982,15 +6109,15 @@ int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask)
5982 if (p->sched_class->set_cpus_allowed) 6109 if (p->sched_class->set_cpus_allowed)
5983 p->sched_class->set_cpus_allowed(p, new_mask); 6110 p->sched_class->set_cpus_allowed(p, new_mask);
5984 else { 6111 else {
5985 p->cpus_allowed = *new_mask; 6112 cpumask_copy(&p->cpus_allowed, new_mask);
5986 p->rt.nr_cpus_allowed = cpus_weight(*new_mask); 6113 p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
5987 } 6114 }
5988 6115
5989 /* Can the task run on the task's current CPU? If so, we're done */ 6116 /* Can the task run on the task's current CPU? If so, we're done */
5990 if (cpu_isset(task_cpu(p), *new_mask)) 6117 if (cpumask_test_cpu(task_cpu(p), new_mask))
5991 goto out; 6118 goto out;
5992 6119
5993 if (migrate_task(p, any_online_cpu(*new_mask), &req)) { 6120 if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) {
5994 /* Need help from migration thread: drop lock and wait. */ 6121 /* Need help from migration thread: drop lock and wait. */
5995 task_rq_unlock(rq, &flags); 6122 task_rq_unlock(rq, &flags);
5996 wake_up_process(rq->migration_thread); 6123 wake_up_process(rq->migration_thread);
@@ -6032,7 +6159,7 @@ static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
6032 if (task_cpu(p) != src_cpu) 6159 if (task_cpu(p) != src_cpu)
6033 goto done; 6160 goto done;
6034 /* Affinity changed (again). */ 6161 /* Affinity changed (again). */
6035 if (!cpu_isset(dest_cpu, p->cpus_allowed)) 6162 if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
6036 goto fail; 6163 goto fail;
6037 6164
6038 on_rq = p->se.on_rq; 6165 on_rq = p->se.on_rq;
@@ -6126,54 +6253,44 @@ static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
6126 6253
6127/* 6254/*
6128 * Figure out where task on dead CPU should go, use force if necessary. 6255 * Figure out where task on dead CPU should go, use force if necessary.
6129 * NOTE: interrupts should be disabled by the caller
6130 */ 6256 */
6131static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) 6257static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
6132{ 6258{
6133 unsigned long flags;
6134 cpumask_t mask;
6135 struct rq *rq;
6136 int dest_cpu; 6259 int dest_cpu;
6260 const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu));
6137 6261
6138 do { 6262again:
6139 /* On same node? */ 6263 /* Look for allowed, online CPU in same node. */
6140 mask = node_to_cpumask(cpu_to_node(dead_cpu)); 6264 for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask)
6141 cpus_and(mask, mask, p->cpus_allowed); 6265 if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
6142 dest_cpu = any_online_cpu(mask); 6266 goto move;
6143 6267
6144 /* On any allowed CPU? */ 6268 /* Any allowed, online CPU? */
6145 if (dest_cpu >= nr_cpu_ids) 6269 dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask);
6146 dest_cpu = any_online_cpu(p->cpus_allowed); 6270 if (dest_cpu < nr_cpu_ids)
6271 goto move;
6147 6272
6148 /* No more Mr. Nice Guy. */ 6273 /* No more Mr. Nice Guy. */
6149 if (dest_cpu >= nr_cpu_ids) { 6274 if (dest_cpu >= nr_cpu_ids) {
6150 cpumask_t cpus_allowed; 6275 cpuset_cpus_allowed_locked(p, &p->cpus_allowed);
6276 dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed);
6151 6277
6152 cpuset_cpus_allowed_locked(p, &cpus_allowed); 6278 /*
6153 /* 6279 * Don't tell them about moving exiting tasks or
6154 * Try to stay on the same cpuset, where the 6280 * kernel threads (both mm NULL), since they never
6155 * current cpuset may be a subset of all cpus. 6281 * leave kernel.
6156 * The cpuset_cpus_allowed_locked() variant of 6282 */
6157 * cpuset_cpus_allowed() will not block. It must be 6283 if (p->mm && printk_ratelimit()) {
6158 * called within calls to cpuset_lock/cpuset_unlock. 6284 printk(KERN_INFO "process %d (%s) no "
6159 */ 6285 "longer affine to cpu%d\n",
6160 rq = task_rq_lock(p, &flags); 6286 task_pid_nr(p), p->comm, dead_cpu);
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 } 6287 }
6176 } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); 6288 }
6289
6290move:
6291 /* It can have affinity changed while we were choosing. */
6292 if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu)))
6293 goto again;
6177} 6294}
6178 6295
6179/* 6296/*
@@ -6185,7 +6302,7 @@ static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
6185 */ 6302 */
6186static void migrate_nr_uninterruptible(struct rq *rq_src) 6303static void migrate_nr_uninterruptible(struct rq *rq_src)
6187{ 6304{
6188 struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); 6305 struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask));
6189 unsigned long flags; 6306 unsigned long flags;
6190 6307
6191 local_irq_save(flags); 6308 local_irq_save(flags);
@@ -6475,7 +6592,7 @@ static void set_rq_online(struct rq *rq)
6475 if (!rq->online) { 6592 if (!rq->online) {
6476 const struct sched_class *class; 6593 const struct sched_class *class;
6477 6594
6478 cpu_set(rq->cpu, rq->rd->online); 6595 cpumask_set_cpu(rq->cpu, rq->rd->online);
6479 rq->online = 1; 6596 rq->online = 1;
6480 6597
6481 for_each_class(class) { 6598 for_each_class(class) {
@@ -6495,7 +6612,7 @@ static void set_rq_offline(struct rq *rq)
6495 class->rq_offline(rq); 6612 class->rq_offline(rq);
6496 } 6613 }
6497 6614
6498 cpu_clear(rq->cpu, rq->rd->online); 6615 cpumask_clear_cpu(rq->cpu, rq->rd->online);
6499 rq->online = 0; 6616 rq->online = 0;
6500 } 6617 }
6501} 6618}
@@ -6536,7 +6653,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6536 rq = cpu_rq(cpu); 6653 rq = cpu_rq(cpu);
6537 spin_lock_irqsave(&rq->lock, flags); 6654 spin_lock_irqsave(&rq->lock, flags);
6538 if (rq->rd) { 6655 if (rq->rd) {
6539 BUG_ON(!cpu_isset(cpu, rq->rd->span)); 6656 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
6540 6657
6541 set_rq_online(rq); 6658 set_rq_online(rq);
6542 } 6659 }
@@ -6550,7 +6667,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6550 break; 6667 break;
6551 /* Unbind it from offline cpu so it can run. Fall thru. */ 6668 /* Unbind it from offline cpu so it can run. Fall thru. */
6552 kthread_bind(cpu_rq(cpu)->migration_thread, 6669 kthread_bind(cpu_rq(cpu)->migration_thread,
6553 any_online_cpu(cpu_online_map)); 6670 cpumask_any(cpu_online_mask));
6554 kthread_stop(cpu_rq(cpu)->migration_thread); 6671 kthread_stop(cpu_rq(cpu)->migration_thread);
6555 cpu_rq(cpu)->migration_thread = NULL; 6672 cpu_rq(cpu)->migration_thread = NULL;
6556 break; 6673 break;
@@ -6587,7 +6704,9 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6587 req = list_entry(rq->migration_queue.next, 6704 req = list_entry(rq->migration_queue.next,
6588 struct migration_req, list); 6705 struct migration_req, list);
6589 list_del_init(&req->list); 6706 list_del_init(&req->list);
6707 spin_unlock_irq(&rq->lock);
6590 complete(&req->done); 6708 complete(&req->done);
6709 spin_lock_irq(&rq->lock);
6591 } 6710 }
6592 spin_unlock_irq(&rq->lock); 6711 spin_unlock_irq(&rq->lock);
6593 break; 6712 break;
@@ -6598,7 +6717,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
6598 rq = cpu_rq(cpu); 6717 rq = cpu_rq(cpu);
6599 spin_lock_irqsave(&rq->lock, flags); 6718 spin_lock_irqsave(&rq->lock, flags);
6600 if (rq->rd) { 6719 if (rq->rd) {
6601 BUG_ON(!cpu_isset(cpu, rq->rd->span)); 6720 BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
6602 set_rq_offline(rq); 6721 set_rq_offline(rq);
6603 } 6722 }
6604 spin_unlock_irqrestore(&rq->lock, flags); 6723 spin_unlock_irqrestore(&rq->lock, flags);
@@ -6636,36 +6755,14 @@ early_initcall(migration_init);
6636 6755
6637#ifdef CONFIG_SCHED_DEBUG 6756#ifdef CONFIG_SCHED_DEBUG
6638 6757
6639static inline const char *sd_level_to_string(enum sched_domain_level lvl)
6640{
6641 switch (lvl) {
6642 case SD_LV_NONE:
6643 return "NONE";
6644 case SD_LV_SIBLING:
6645 return "SIBLING";
6646 case SD_LV_MC:
6647 return "MC";
6648 case SD_LV_CPU:
6649 return "CPU";
6650 case SD_LV_NODE:
6651 return "NODE";
6652 case SD_LV_ALLNODES:
6653 return "ALLNODES";
6654 case SD_LV_MAX:
6655 return "MAX";
6656
6657 }
6658 return "MAX";
6659}
6660
6661static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, 6758static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6662 cpumask_t *groupmask) 6759 struct cpumask *groupmask)
6663{ 6760{
6664 struct sched_group *group = sd->groups; 6761 struct sched_group *group = sd->groups;
6665 char str[256]; 6762 char str[256];
6666 6763
6667 cpulist_scnprintf(str, sizeof(str), sd->span); 6764 cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
6668 cpus_clear(*groupmask); 6765 cpumask_clear(groupmask);
6669 6766
6670 printk(KERN_DEBUG "%*s domain %d: ", level, "", level); 6767 printk(KERN_DEBUG "%*s domain %d: ", level, "", level);
6671 6768
@@ -6677,14 +6774,13 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6677 return -1; 6774 return -1;
6678 } 6775 }
6679 6776
6680 printk(KERN_CONT "span %s level %s\n", 6777 printk(KERN_CONT "span %s level %s\n", str, sd->name);
6681 str, sd_level_to_string(sd->level));
6682 6778
6683 if (!cpu_isset(cpu, sd->span)) { 6779 if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
6684 printk(KERN_ERR "ERROR: domain->span does not contain " 6780 printk(KERN_ERR "ERROR: domain->span does not contain "
6685 "CPU%d\n", cpu); 6781 "CPU%d\n", cpu);
6686 } 6782 }
6687 if (!cpu_isset(cpu, group->cpumask)) { 6783 if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
6688 printk(KERN_ERR "ERROR: domain->groups does not contain" 6784 printk(KERN_ERR "ERROR: domain->groups does not contain"
6689 " CPU%d\n", cpu); 6785 " CPU%d\n", cpu);
6690 } 6786 }
@@ -6704,31 +6800,32 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6704 break; 6800 break;
6705 } 6801 }
6706 6802
6707 if (!cpus_weight(group->cpumask)) { 6803 if (!cpumask_weight(sched_group_cpus(group))) {
6708 printk(KERN_CONT "\n"); 6804 printk(KERN_CONT "\n");
6709 printk(KERN_ERR "ERROR: empty group\n"); 6805 printk(KERN_ERR "ERROR: empty group\n");
6710 break; 6806 break;
6711 } 6807 }
6712 6808
6713 if (cpus_intersects(*groupmask, group->cpumask)) { 6809 if (cpumask_intersects(groupmask, sched_group_cpus(group))) {
6714 printk(KERN_CONT "\n"); 6810 printk(KERN_CONT "\n");
6715 printk(KERN_ERR "ERROR: repeated CPUs\n"); 6811 printk(KERN_ERR "ERROR: repeated CPUs\n");
6716 break; 6812 break;
6717 } 6813 }
6718 6814
6719 cpus_or(*groupmask, *groupmask, group->cpumask); 6815 cpumask_or(groupmask, groupmask, sched_group_cpus(group));
6720 6816
6721 cpulist_scnprintf(str, sizeof(str), group->cpumask); 6817 cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
6722 printk(KERN_CONT " %s", str); 6818 printk(KERN_CONT " %s", str);
6723 6819
6724 group = group->next; 6820 group = group->next;
6725 } while (group != sd->groups); 6821 } while (group != sd->groups);
6726 printk(KERN_CONT "\n"); 6822 printk(KERN_CONT "\n");
6727 6823
6728 if (!cpus_equal(sd->span, *groupmask)) 6824 if (!cpumask_equal(sched_domain_span(sd), groupmask))
6729 printk(KERN_ERR "ERROR: groups don't span domain->span\n"); 6825 printk(KERN_ERR "ERROR: groups don't span domain->span\n");
6730 6826
6731 if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) 6827 if (sd->parent &&
6828 !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
6732 printk(KERN_ERR "ERROR: parent span is not a superset " 6829 printk(KERN_ERR "ERROR: parent span is not a superset "
6733 "of domain->span\n"); 6830 "of domain->span\n");
6734 return 0; 6831 return 0;
@@ -6736,7 +6833,7 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
6736 6833
6737static void sched_domain_debug(struct sched_domain *sd, int cpu) 6834static void sched_domain_debug(struct sched_domain *sd, int cpu)
6738{ 6835{
6739 cpumask_t *groupmask; 6836 cpumask_var_t groupmask;
6740 int level = 0; 6837 int level = 0;
6741 6838
6742 if (!sd) { 6839 if (!sd) {
@@ -6746,8 +6843,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
6746 6843
6747 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); 6844 printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);
6748 6845
6749 groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); 6846 if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) {
6750 if (!groupmask) {
6751 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); 6847 printk(KERN_DEBUG "Cannot load-balance (out of memory)\n");
6752 return; 6848 return;
6753 } 6849 }
@@ -6760,7 +6856,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
6760 if (!sd) 6856 if (!sd)
6761 break; 6857 break;
6762 } 6858 }
6763 kfree(groupmask); 6859 free_cpumask_var(groupmask);
6764} 6860}
6765#else /* !CONFIG_SCHED_DEBUG */ 6861#else /* !CONFIG_SCHED_DEBUG */
6766# define sched_domain_debug(sd, cpu) do { } while (0) 6862# define sched_domain_debug(sd, cpu) do { } while (0)
@@ -6768,7 +6864,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
6768 6864
6769static int sd_degenerate(struct sched_domain *sd) 6865static int sd_degenerate(struct sched_domain *sd)
6770{ 6866{
6771 if (cpus_weight(sd->span) == 1) 6867 if (cpumask_weight(sched_domain_span(sd)) == 1)
6772 return 1; 6868 return 1;
6773 6869
6774 /* Following flags need at least 2 groups */ 6870 /* Following flags need at least 2 groups */
@@ -6799,7 +6895,7 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6799 if (sd_degenerate(parent)) 6895 if (sd_degenerate(parent))
6800 return 1; 6896 return 1;
6801 6897
6802 if (!cpus_equal(sd->span, parent->span)) 6898 if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
6803 return 0; 6899 return 0;
6804 6900
6805 /* Does parent contain flags not in child? */ 6901 /* Does parent contain flags not in child? */
@@ -6814,6 +6910,8 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6814 SD_BALANCE_EXEC | 6910 SD_BALANCE_EXEC |
6815 SD_SHARE_CPUPOWER | 6911 SD_SHARE_CPUPOWER |
6816 SD_SHARE_PKG_RESOURCES); 6912 SD_SHARE_PKG_RESOURCES);
6913 if (nr_node_ids == 1)
6914 pflags &= ~SD_SERIALIZE;
6817 } 6915 }
6818 if (~cflags & pflags) 6916 if (~cflags & pflags)
6819 return 0; 6917 return 0;
@@ -6821,6 +6919,16 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
6821 return 1; 6919 return 1;
6822} 6920}
6823 6921
6922static void free_rootdomain(struct root_domain *rd)
6923{
6924 cpupri_cleanup(&rd->cpupri);
6925
6926 free_cpumask_var(rd->rto_mask);
6927 free_cpumask_var(rd->online);
6928 free_cpumask_var(rd->span);
6929 kfree(rd);
6930}
6931
6824static void rq_attach_root(struct rq *rq, struct root_domain *rd) 6932static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6825{ 6933{
6826 unsigned long flags; 6934 unsigned long flags;
@@ -6830,38 +6938,63 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd)
6830 if (rq->rd) { 6938 if (rq->rd) {
6831 struct root_domain *old_rd = rq->rd; 6939 struct root_domain *old_rd = rq->rd;
6832 6940
6833 if (cpu_isset(rq->cpu, old_rd->online)) 6941 if (cpumask_test_cpu(rq->cpu, old_rd->online))
6834 set_rq_offline(rq); 6942 set_rq_offline(rq);
6835 6943
6836 cpu_clear(rq->cpu, old_rd->span); 6944 cpumask_clear_cpu(rq->cpu, old_rd->span);
6837 6945
6838 if (atomic_dec_and_test(&old_rd->refcount)) 6946 if (atomic_dec_and_test(&old_rd->refcount))
6839 kfree(old_rd); 6947 free_rootdomain(old_rd);
6840 } 6948 }
6841 6949
6842 atomic_inc(&rd->refcount); 6950 atomic_inc(&rd->refcount);
6843 rq->rd = rd; 6951 rq->rd = rd;
6844 6952
6845 cpu_set(rq->cpu, rd->span); 6953 cpumask_set_cpu(rq->cpu, rd->span);
6846 if (cpu_isset(rq->cpu, cpu_online_map)) 6954 if (cpumask_test_cpu(rq->cpu, cpu_online_mask))
6847 set_rq_online(rq); 6955 set_rq_online(rq);
6848 6956
6849 spin_unlock_irqrestore(&rq->lock, flags); 6957 spin_unlock_irqrestore(&rq->lock, flags);
6850} 6958}
6851 6959
6852static void init_rootdomain(struct root_domain *rd) 6960static int init_rootdomain(struct root_domain *rd, bool bootmem)
6853{ 6961{
6854 memset(rd, 0, sizeof(*rd)); 6962 memset(rd, 0, sizeof(*rd));
6855 6963
6856 cpus_clear(rd->span); 6964 if (bootmem) {
6857 cpus_clear(rd->online); 6965 alloc_bootmem_cpumask_var(&def_root_domain.span);
6966 alloc_bootmem_cpumask_var(&def_root_domain.online);
6967 alloc_bootmem_cpumask_var(&def_root_domain.rto_mask);
6968 cpupri_init(&rd->cpupri, true);
6969 return 0;
6970 }
6858 6971
6859 cpupri_init(&rd->cpupri); 6972 if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
6973 goto free_rd;
6974 if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
6975 goto free_span;
6976 if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
6977 goto free_online;
6978
6979 if (cpupri_init(&rd->cpupri, false) != 0)
6980 goto free_rto_mask;
6981 return 0;
6982
6983free_rto_mask:
6984 free_cpumask_var(rd->rto_mask);
6985free_online:
6986 free_cpumask_var(rd->online);
6987free_span:
6988 free_cpumask_var(rd->span);
6989free_rd:
6990 kfree(rd);
6991 return -ENOMEM;
6860} 6992}
6861 6993
6862static void init_defrootdomain(void) 6994static void init_defrootdomain(void)
6863{ 6995{
6864 init_rootdomain(&def_root_domain); 6996 init_rootdomain(&def_root_domain, true);
6997
6865 atomic_set(&def_root_domain.refcount, 1); 6998 atomic_set(&def_root_domain.refcount, 1);
6866} 6999}
6867 7000
@@ -6873,7 +7006,10 @@ static struct root_domain *alloc_rootdomain(void)
6873 if (!rd) 7006 if (!rd)
6874 return NULL; 7007 return NULL;
6875 7008
6876 init_rootdomain(rd); 7009 if (init_rootdomain(rd, false) != 0) {
7010 kfree(rd);
7011 return NULL;
7012 }
6877 7013
6878 return rd; 7014 return rd;
6879} 7015}
@@ -6915,19 +7051,12 @@ cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
6915} 7051}
6916 7052
6917/* cpus with isolated domains */ 7053/* cpus with isolated domains */
6918static cpumask_t cpu_isolated_map = CPU_MASK_NONE; 7054static cpumask_var_t cpu_isolated_map;
6919 7055
6920/* Setup the mask of cpus configured for isolated domains */ 7056/* Setup the mask of cpus configured for isolated domains */
6921static int __init isolated_cpu_setup(char *str) 7057static int __init isolated_cpu_setup(char *str)
6922{ 7058{
6923 static int __initdata ints[NR_CPUS]; 7059 cpulist_parse(str, cpu_isolated_map);
6924 int i;
6925
6926 str = get_options(str, ARRAY_SIZE(ints), ints);
6927 cpus_clear(cpu_isolated_map);
6928 for (i = 1; i <= ints[0]; i++)
6929 if (ints[i] < NR_CPUS)
6930 cpu_set(ints[i], cpu_isolated_map);
6931 return 1; 7060 return 1;
6932} 7061}
6933 7062
@@ -6936,42 +7065,43 @@ __setup("isolcpus=", isolated_cpu_setup);
6936/* 7065/*
6937 * init_sched_build_groups takes the cpumask we wish to span, and a pointer 7066 * init_sched_build_groups takes the cpumask we wish to span, and a pointer
6938 * to a function which identifies what group(along with sched group) a CPU 7067 * to a function which identifies what group(along with sched group) a CPU
6939 * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS 7068 * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
6940 * (due to the fact that we keep track of groups covered with a cpumask_t). 7069 * (due to the fact that we keep track of groups covered with a struct cpumask).
6941 * 7070 *
6942 * init_sched_build_groups will build a circular linked list of the groups 7071 * init_sched_build_groups will build a circular linked list of the groups
6943 * covered by the given span, and will set each group's ->cpumask correctly, 7072 * covered by the given span, and will set each group's ->cpumask correctly,
6944 * and ->cpu_power to 0. 7073 * and ->cpu_power to 0.
6945 */ 7074 */
6946static void 7075static void
6947init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, 7076init_sched_build_groups(const struct cpumask *span,
6948 int (*group_fn)(int cpu, const cpumask_t *cpu_map, 7077 const struct cpumask *cpu_map,
7078 int (*group_fn)(int cpu, const struct cpumask *cpu_map,
6949 struct sched_group **sg, 7079 struct sched_group **sg,
6950 cpumask_t *tmpmask), 7080 struct cpumask *tmpmask),
6951 cpumask_t *covered, cpumask_t *tmpmask) 7081 struct cpumask *covered, struct cpumask *tmpmask)
6952{ 7082{
6953 struct sched_group *first = NULL, *last = NULL; 7083 struct sched_group *first = NULL, *last = NULL;
6954 int i; 7084 int i;
6955 7085
6956 cpus_clear(*covered); 7086 cpumask_clear(covered);
6957 7087
6958 for_each_cpu_mask_nr(i, *span) { 7088 for_each_cpu(i, span) {
6959 struct sched_group *sg; 7089 struct sched_group *sg;
6960 int group = group_fn(i, cpu_map, &sg, tmpmask); 7090 int group = group_fn(i, cpu_map, &sg, tmpmask);
6961 int j; 7091 int j;
6962 7092
6963 if (cpu_isset(i, *covered)) 7093 if (cpumask_test_cpu(i, covered))
6964 continue; 7094 continue;
6965 7095
6966 cpus_clear(sg->cpumask); 7096 cpumask_clear(sched_group_cpus(sg));
6967 sg->__cpu_power = 0; 7097 sg->__cpu_power = 0;
6968 7098
6969 for_each_cpu_mask_nr(j, *span) { 7099 for_each_cpu(j, span) {
6970 if (group_fn(j, cpu_map, NULL, tmpmask) != group) 7100 if (group_fn(j, cpu_map, NULL, tmpmask) != group)
6971 continue; 7101 continue;
6972 7102
6973 cpu_set(j, *covered); 7103 cpumask_set_cpu(j, covered);
6974 cpu_set(j, sg->cpumask); 7104 cpumask_set_cpu(j, sched_group_cpus(sg));
6975 } 7105 }
6976 if (!first) 7106 if (!first)
6977 first = sg; 7107 first = sg;
@@ -7035,23 +7165,21 @@ static int find_next_best_node(int node, nodemask_t *used_nodes)
7035 * should be one that prevents unnecessary balancing, but also spreads tasks 7165 * should be one that prevents unnecessary balancing, but also spreads tasks
7036 * out optimally. 7166 * out optimally.
7037 */ 7167 */
7038static void sched_domain_node_span(int node, cpumask_t *span) 7168static void sched_domain_node_span(int node, struct cpumask *span)
7039{ 7169{
7040 nodemask_t used_nodes; 7170 nodemask_t used_nodes;
7041 node_to_cpumask_ptr(nodemask, node);
7042 int i; 7171 int i;
7043 7172
7044 cpus_clear(*span); 7173 cpumask_clear(span);
7045 nodes_clear(used_nodes); 7174 nodes_clear(used_nodes);
7046 7175
7047 cpus_or(*span, *span, *nodemask); 7176 cpumask_or(span, span, cpumask_of_node(node));
7048 node_set(node, used_nodes); 7177 node_set(node, used_nodes);
7049 7178
7050 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { 7179 for (i = 1; i < SD_NODES_PER_DOMAIN; i++) {
7051 int next_node = find_next_best_node(node, &used_nodes); 7180 int next_node = find_next_best_node(node, &used_nodes);
7052 7181
7053 node_to_cpumask_ptr_next(nodemask, next_node); 7182 cpumask_or(span, span, cpumask_of_node(next_node));
7054 cpus_or(*span, *span, *nodemask);
7055 } 7183 }
7056} 7184}
7057#endif /* CONFIG_NUMA */ 7185#endif /* CONFIG_NUMA */
@@ -7059,18 +7187,33 @@ static void sched_domain_node_span(int node, cpumask_t *span)
7059int sched_smt_power_savings = 0, sched_mc_power_savings = 0; 7187int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
7060 7188
7061/* 7189/*
7190 * The cpus mask in sched_group and sched_domain hangs off the end.
7191 * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space
7192 * for nr_cpu_ids < CONFIG_NR_CPUS.
7193 */
7194struct static_sched_group {
7195 struct sched_group sg;
7196 DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
7197};
7198
7199struct static_sched_domain {
7200 struct sched_domain sd;
7201 DECLARE_BITMAP(span, CONFIG_NR_CPUS);
7202};
7203
7204/*
7062 * SMT sched-domains: 7205 * SMT sched-domains:
7063 */ 7206 */
7064#ifdef CONFIG_SCHED_SMT 7207#ifdef CONFIG_SCHED_SMT
7065static DEFINE_PER_CPU(struct sched_domain, cpu_domains); 7208static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
7066static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); 7209static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus);
7067 7210
7068static int 7211static int
7069cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7212cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
7070 cpumask_t *unused) 7213 struct sched_group **sg, struct cpumask *unused)
7071{ 7214{
7072 if (sg) 7215 if (sg)
7073 *sg = &per_cpu(sched_group_cpus, cpu); 7216 *sg = &per_cpu(sched_group_cpus, cpu).sg;
7074 return cpu; 7217 return cpu;
7075} 7218}
7076#endif /* CONFIG_SCHED_SMT */ 7219#endif /* CONFIG_SCHED_SMT */
@@ -7079,56 +7222,53 @@ cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg,
7079 * multi-core sched-domains: 7222 * multi-core sched-domains:
7080 */ 7223 */
7081#ifdef CONFIG_SCHED_MC 7224#ifdef CONFIG_SCHED_MC
7082static DEFINE_PER_CPU(struct sched_domain, core_domains); 7225static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
7083static DEFINE_PER_CPU(struct sched_group, sched_group_core); 7226static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
7084#endif /* CONFIG_SCHED_MC */ 7227#endif /* CONFIG_SCHED_MC */
7085 7228
7086#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) 7229#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
7087static int 7230static int
7088cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7231cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
7089 cpumask_t *mask) 7232 struct sched_group **sg, struct cpumask *mask)
7090{ 7233{
7091 int group; 7234 int group;
7092 7235
7093 *mask = per_cpu(cpu_sibling_map, cpu); 7236 cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map);
7094 cpus_and(*mask, *mask, *cpu_map); 7237 group = cpumask_first(mask);
7095 group = first_cpu(*mask);
7096 if (sg) 7238 if (sg)
7097 *sg = &per_cpu(sched_group_core, group); 7239 *sg = &per_cpu(sched_group_core, group).sg;
7098 return group; 7240 return group;
7099} 7241}
7100#elif defined(CONFIG_SCHED_MC) 7242#elif defined(CONFIG_SCHED_MC)
7101static int 7243static int
7102cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7244cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
7103 cpumask_t *unused) 7245 struct sched_group **sg, struct cpumask *unused)
7104{ 7246{
7105 if (sg) 7247 if (sg)
7106 *sg = &per_cpu(sched_group_core, cpu); 7248 *sg = &per_cpu(sched_group_core, cpu).sg;
7107 return cpu; 7249 return cpu;
7108} 7250}
7109#endif 7251#endif
7110 7252
7111static DEFINE_PER_CPU(struct sched_domain, phys_domains); 7253static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
7112static DEFINE_PER_CPU(struct sched_group, sched_group_phys); 7254static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
7113 7255
7114static int 7256static int
7115cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, 7257cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
7116 cpumask_t *mask) 7258 struct sched_group **sg, struct cpumask *mask)
7117{ 7259{
7118 int group; 7260 int group;
7119#ifdef CONFIG_SCHED_MC 7261#ifdef CONFIG_SCHED_MC
7120 *mask = cpu_coregroup_map(cpu); 7262 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
7121 cpus_and(*mask, *mask, *cpu_map); 7263 group = cpumask_first(mask);
7122 group = first_cpu(*mask);
7123#elif defined(CONFIG_SCHED_SMT) 7264#elif defined(CONFIG_SCHED_SMT)
7124 *mask = per_cpu(cpu_sibling_map, cpu); 7265 cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map);
7125 cpus_and(*mask, *mask, *cpu_map); 7266 group = cpumask_first(mask);
7126 group = first_cpu(*mask);
7127#else 7267#else
7128 group = cpu; 7268 group = cpu;
7129#endif 7269#endif
7130 if (sg) 7270 if (sg)
7131 *sg = &per_cpu(sched_group_phys, group); 7271 *sg = &per_cpu(sched_group_phys, group).sg;
7132 return group; 7272 return group;
7133} 7273}
7134 7274
@@ -7142,19 +7282,19 @@ static DEFINE_PER_CPU(struct sched_domain, node_domains);
7142static struct sched_group ***sched_group_nodes_bycpu; 7282static struct sched_group ***sched_group_nodes_bycpu;
7143 7283
7144static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); 7284static DEFINE_PER_CPU(struct sched_domain, allnodes_domains);
7145static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); 7285static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
7146 7286
7147static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, 7287static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
7148 struct sched_group **sg, cpumask_t *nodemask) 7288 struct sched_group **sg,
7289 struct cpumask *nodemask)
7149{ 7290{
7150 int group; 7291 int group;
7151 7292
7152 *nodemask = node_to_cpumask(cpu_to_node(cpu)); 7293 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
7153 cpus_and(*nodemask, *nodemask, *cpu_map); 7294 group = cpumask_first(nodemask);
7154 group = first_cpu(*nodemask);
7155 7295
7156 if (sg) 7296 if (sg)
7157 *sg = &per_cpu(sched_group_allnodes, group); 7297 *sg = &per_cpu(sched_group_allnodes, group).sg;
7158 return group; 7298 return group;
7159} 7299}
7160 7300
@@ -7166,11 +7306,11 @@ static void init_numa_sched_groups_power(struct sched_group *group_head)
7166 if (!sg) 7306 if (!sg)
7167 return; 7307 return;
7168 do { 7308 do {
7169 for_each_cpu_mask_nr(j, sg->cpumask) { 7309 for_each_cpu(j, sched_group_cpus(sg)) {
7170 struct sched_domain *sd; 7310 struct sched_domain *sd;
7171 7311
7172 sd = &per_cpu(phys_domains, j); 7312 sd = &per_cpu(phys_domains, j).sd;
7173 if (j != first_cpu(sd->groups->cpumask)) { 7313 if (j != cpumask_first(sched_group_cpus(sd->groups))) {
7174 /* 7314 /*
7175 * Only add "power" once for each 7315 * Only add "power" once for each
7176 * physical package. 7316 * physical package.
@@ -7187,11 +7327,12 @@ static void init_numa_sched_groups_power(struct sched_group *group_head)
7187 7327
7188#ifdef CONFIG_NUMA 7328#ifdef CONFIG_NUMA
7189/* Free memory allocated for various sched_group structures */ 7329/* Free memory allocated for various sched_group structures */
7190static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) 7330static void free_sched_groups(const struct cpumask *cpu_map,
7331 struct cpumask *nodemask)
7191{ 7332{
7192 int cpu, i; 7333 int cpu, i;
7193 7334
7194 for_each_cpu_mask_nr(cpu, *cpu_map) { 7335 for_each_cpu(cpu, cpu_map) {
7195 struct sched_group **sched_group_nodes 7336 struct sched_group **sched_group_nodes
7196 = sched_group_nodes_bycpu[cpu]; 7337 = sched_group_nodes_bycpu[cpu];
7197 7338
@@ -7201,9 +7342,8 @@ static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask)
7201 for (i = 0; i < nr_node_ids; i++) { 7342 for (i = 0; i < nr_node_ids; i++) {
7202 struct sched_group *oldsg, *sg = sched_group_nodes[i]; 7343 struct sched_group *oldsg, *sg = sched_group_nodes[i];
7203 7344
7204 *nodemask = node_to_cpumask(i); 7345 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
7205 cpus_and(*nodemask, *nodemask, *cpu_map); 7346 if (cpumask_empty(nodemask))
7206 if (cpus_empty(*nodemask))
7207 continue; 7347 continue;
7208 7348
7209 if (sg == NULL) 7349 if (sg == NULL)
@@ -7221,7 +7361,8 @@ next_sg:
7221 } 7361 }
7222} 7362}
7223#else /* !CONFIG_NUMA */ 7363#else /* !CONFIG_NUMA */
7224static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) 7364static void free_sched_groups(const struct cpumask *cpu_map,
7365 struct cpumask *nodemask)
7225{ 7366{
7226} 7367}
7227#endif /* CONFIG_NUMA */ 7368#endif /* CONFIG_NUMA */
@@ -7247,7 +7388,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
7247 7388
7248 WARN_ON(!sd || !sd->groups); 7389 WARN_ON(!sd || !sd->groups);
7249 7390
7250 if (cpu != first_cpu(sd->groups->cpumask)) 7391 if (cpu != cpumask_first(sched_group_cpus(sd->groups)))
7251 return; 7392 return;
7252 7393
7253 child = sd->child; 7394 child = sd->child;
@@ -7312,40 +7453,6 @@ SD_INIT_FUNC(CPU)
7312 SD_INIT_FUNC(MC) 7453 SD_INIT_FUNC(MC)
7313#endif 7454#endif
7314 7455
7315/*
7316 * To minimize stack usage kmalloc room for cpumasks and share the
7317 * space as the usage in build_sched_domains() dictates. Used only
7318 * if the amount of space is significant.
7319 */
7320struct allmasks {
7321 cpumask_t tmpmask; /* make this one first */
7322 union {
7323 cpumask_t nodemask;
7324 cpumask_t this_sibling_map;
7325 cpumask_t this_core_map;
7326 };
7327 cpumask_t send_covered;
7328
7329#ifdef CONFIG_NUMA
7330 cpumask_t domainspan;
7331 cpumask_t covered;
7332 cpumask_t notcovered;
7333#endif
7334};
7335
7336#if NR_CPUS > 128
7337#define SCHED_CPUMASK_ALLOC 1
7338#define SCHED_CPUMASK_FREE(v) kfree(v)
7339#define SCHED_CPUMASK_DECLARE(v) struct allmasks *v
7340#else
7341#define SCHED_CPUMASK_ALLOC 0
7342#define SCHED_CPUMASK_FREE(v)
7343#define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v
7344#endif
7345
7346#define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \
7347 ((unsigned long)(a) + offsetof(struct allmasks, v))
7348
7349static int default_relax_domain_level = -1; 7456static int default_relax_domain_level = -1;
7350 7457
7351static int __init setup_relax_domain_level(char *str) 7458static int __init setup_relax_domain_level(char *str)
@@ -7385,17 +7492,38 @@ static void set_domain_attribute(struct sched_domain *sd,
7385 * Build sched domains for a given set of cpus and attach the sched domains 7492 * Build sched domains for a given set of cpus and attach the sched domains
7386 * to the individual cpus 7493 * to the individual cpus
7387 */ 7494 */
7388static int __build_sched_domains(const cpumask_t *cpu_map, 7495static int __build_sched_domains(const struct cpumask *cpu_map,
7389 struct sched_domain_attr *attr) 7496 struct sched_domain_attr *attr)
7390{ 7497{
7391 int i; 7498 int i, err = -ENOMEM;
7392 struct root_domain *rd; 7499 struct root_domain *rd;
7393 SCHED_CPUMASK_DECLARE(allmasks); 7500 cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered,
7394 cpumask_t *tmpmask; 7501 tmpmask;
7395#ifdef CONFIG_NUMA 7502#ifdef CONFIG_NUMA
7503 cpumask_var_t domainspan, covered, notcovered;
7396 struct sched_group **sched_group_nodes = NULL; 7504 struct sched_group **sched_group_nodes = NULL;
7397 int sd_allnodes = 0; 7505 int sd_allnodes = 0;
7398 7506
7507 if (!alloc_cpumask_var(&domainspan, GFP_KERNEL))
7508 goto out;
7509 if (!alloc_cpumask_var(&covered, GFP_KERNEL))
7510 goto free_domainspan;
7511 if (!alloc_cpumask_var(&notcovered, GFP_KERNEL))
7512 goto free_covered;
7513#endif
7514
7515 if (!alloc_cpumask_var(&nodemask, GFP_KERNEL))
7516 goto free_notcovered;
7517 if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL))
7518 goto free_nodemask;
7519 if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL))
7520 goto free_this_sibling_map;
7521 if (!alloc_cpumask_var(&send_covered, GFP_KERNEL))
7522 goto free_this_core_map;
7523 if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL))
7524 goto free_send_covered;
7525
7526#ifdef CONFIG_NUMA
7399 /* 7527 /*
7400 * Allocate the per-node list of sched groups 7528 * Allocate the per-node list of sched groups
7401 */ 7529 */
@@ -7403,55 +7531,35 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7403 GFP_KERNEL); 7531 GFP_KERNEL);
7404 if (!sched_group_nodes) { 7532 if (!sched_group_nodes) {
7405 printk(KERN_WARNING "Can not alloc sched group node list\n"); 7533 printk(KERN_WARNING "Can not alloc sched group node list\n");
7406 return -ENOMEM; 7534 goto free_tmpmask;
7407 } 7535 }
7408#endif 7536#endif
7409 7537
7410 rd = alloc_rootdomain(); 7538 rd = alloc_rootdomain();
7411 if (!rd) { 7539 if (!rd) {
7412 printk(KERN_WARNING "Cannot alloc root domain\n"); 7540 printk(KERN_WARNING "Cannot alloc root domain\n");
7413#ifdef CONFIG_NUMA 7541 goto free_sched_groups;
7414 kfree(sched_group_nodes);
7415#endif
7416 return -ENOMEM;
7417 } 7542 }
7418 7543
7419#if SCHED_CPUMASK_ALLOC
7420 /* get space for all scratch cpumask variables */
7421 allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL);
7422 if (!allmasks) {
7423 printk(KERN_WARNING "Cannot alloc cpumask array\n");
7424 kfree(rd);
7425#ifdef CONFIG_NUMA 7544#ifdef CONFIG_NUMA
7426 kfree(sched_group_nodes); 7545 sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes;
7427#endif
7428 return -ENOMEM;
7429 }
7430#endif
7431 tmpmask = (cpumask_t *)allmasks;
7432
7433
7434#ifdef CONFIG_NUMA
7435 sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes;
7436#endif 7546#endif
7437 7547
7438 /* 7548 /*
7439 * Set up domains for cpus specified by the cpu_map. 7549 * Set up domains for cpus specified by the cpu_map.
7440 */ 7550 */
7441 for_each_cpu_mask_nr(i, *cpu_map) { 7551 for_each_cpu(i, cpu_map) {
7442 struct sched_domain *sd = NULL, *p; 7552 struct sched_domain *sd = NULL, *p;
7443 SCHED_CPUMASK_VAR(nodemask, allmasks);
7444 7553
7445 *nodemask = node_to_cpumask(cpu_to_node(i)); 7554 cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map);
7446 cpus_and(*nodemask, *nodemask, *cpu_map);
7447 7555
7448#ifdef CONFIG_NUMA 7556#ifdef CONFIG_NUMA
7449 if (cpus_weight(*cpu_map) > 7557 if (cpumask_weight(cpu_map) >
7450 SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { 7558 SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) {
7451 sd = &per_cpu(allnodes_domains, i); 7559 sd = &per_cpu(allnodes_domains, i);
7452 SD_INIT(sd, ALLNODES); 7560 SD_INIT(sd, ALLNODES);
7453 set_domain_attribute(sd, attr); 7561 set_domain_attribute(sd, attr);
7454 sd->span = *cpu_map; 7562 cpumask_copy(sched_domain_span(sd), cpu_map);
7455 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); 7563 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask);
7456 p = sd; 7564 p = sd;
7457 sd_allnodes = 1; 7565 sd_allnodes = 1;
@@ -7461,18 +7569,19 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7461 sd = &per_cpu(node_domains, i); 7569 sd = &per_cpu(node_domains, i);
7462 SD_INIT(sd, NODE); 7570 SD_INIT(sd, NODE);
7463 set_domain_attribute(sd, attr); 7571 set_domain_attribute(sd, attr);
7464 sched_domain_node_span(cpu_to_node(i), &sd->span); 7572 sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
7465 sd->parent = p; 7573 sd->parent = p;
7466 if (p) 7574 if (p)
7467 p->child = sd; 7575 p->child = sd;
7468 cpus_and(sd->span, sd->span, *cpu_map); 7576 cpumask_and(sched_domain_span(sd),
7577 sched_domain_span(sd), cpu_map);
7469#endif 7578#endif
7470 7579
7471 p = sd; 7580 p = sd;
7472 sd = &per_cpu(phys_domains, i); 7581 sd = &per_cpu(phys_domains, i).sd;
7473 SD_INIT(sd, CPU); 7582 SD_INIT(sd, CPU);
7474 set_domain_attribute(sd, attr); 7583 set_domain_attribute(sd, attr);
7475 sd->span = *nodemask; 7584 cpumask_copy(sched_domain_span(sd), nodemask);
7476 sd->parent = p; 7585 sd->parent = p;
7477 if (p) 7586 if (p)
7478 p->child = sd; 7587 p->child = sd;
@@ -7480,11 +7589,11 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7480 7589
7481#ifdef CONFIG_SCHED_MC 7590#ifdef CONFIG_SCHED_MC
7482 p = sd; 7591 p = sd;
7483 sd = &per_cpu(core_domains, i); 7592 sd = &per_cpu(core_domains, i).sd;
7484 SD_INIT(sd, MC); 7593 SD_INIT(sd, MC);
7485 set_domain_attribute(sd, attr); 7594 set_domain_attribute(sd, attr);
7486 sd->span = cpu_coregroup_map(i); 7595 cpumask_and(sched_domain_span(sd), cpu_map,
7487 cpus_and(sd->span, sd->span, *cpu_map); 7596 cpu_coregroup_mask(i));
7488 sd->parent = p; 7597 sd->parent = p;
7489 p->child = sd; 7598 p->child = sd;
7490 cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); 7599 cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask);
@@ -7492,11 +7601,11 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7492 7601
7493#ifdef CONFIG_SCHED_SMT 7602#ifdef CONFIG_SCHED_SMT
7494 p = sd; 7603 p = sd;
7495 sd = &per_cpu(cpu_domains, i); 7604 sd = &per_cpu(cpu_domains, i).sd;
7496 SD_INIT(sd, SIBLING); 7605 SD_INIT(sd, SIBLING);
7497 set_domain_attribute(sd, attr); 7606 set_domain_attribute(sd, attr);
7498 sd->span = per_cpu(cpu_sibling_map, i); 7607 cpumask_and(sched_domain_span(sd),
7499 cpus_and(sd->span, sd->span, *cpu_map); 7608 &per_cpu(cpu_sibling_map, i), cpu_map);
7500 sd->parent = p; 7609 sd->parent = p;
7501 p->child = sd; 7610 p->child = sd;
7502 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); 7611 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask);
@@ -7505,13 +7614,10 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7505 7614
7506#ifdef CONFIG_SCHED_SMT 7615#ifdef CONFIG_SCHED_SMT
7507 /* Set up CPU (sibling) groups */ 7616 /* Set up CPU (sibling) groups */
7508 for_each_cpu_mask_nr(i, *cpu_map) { 7617 for_each_cpu(i, cpu_map) {
7509 SCHED_CPUMASK_VAR(this_sibling_map, allmasks); 7618 cpumask_and(this_sibling_map,
7510 SCHED_CPUMASK_VAR(send_covered, allmasks); 7619 &per_cpu(cpu_sibling_map, i), cpu_map);
7511 7620 if (i != cpumask_first(this_sibling_map))
7512 *this_sibling_map = per_cpu(cpu_sibling_map, i);
7513 cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map);
7514 if (i != first_cpu(*this_sibling_map))
7515 continue; 7621 continue;
7516 7622
7517 init_sched_build_groups(this_sibling_map, cpu_map, 7623 init_sched_build_groups(this_sibling_map, cpu_map,
@@ -7522,13 +7628,9 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7522 7628
7523#ifdef CONFIG_SCHED_MC 7629#ifdef CONFIG_SCHED_MC
7524 /* Set up multi-core groups */ 7630 /* Set up multi-core groups */
7525 for_each_cpu_mask_nr(i, *cpu_map) { 7631 for_each_cpu(i, cpu_map) {
7526 SCHED_CPUMASK_VAR(this_core_map, allmasks); 7632 cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map);
7527 SCHED_CPUMASK_VAR(send_covered, allmasks); 7633 if (i != cpumask_first(this_core_map))
7528
7529 *this_core_map = cpu_coregroup_map(i);
7530 cpus_and(*this_core_map, *this_core_map, *cpu_map);
7531 if (i != first_cpu(*this_core_map))
7532 continue; 7634 continue;
7533 7635
7534 init_sched_build_groups(this_core_map, cpu_map, 7636 init_sched_build_groups(this_core_map, cpu_map,
@@ -7539,12 +7641,8 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7539 7641
7540 /* Set up physical groups */ 7642 /* Set up physical groups */
7541 for (i = 0; i < nr_node_ids; i++) { 7643 for (i = 0; i < nr_node_ids; i++) {
7542 SCHED_CPUMASK_VAR(nodemask, allmasks); 7644 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
7543 SCHED_CPUMASK_VAR(send_covered, allmasks); 7645 if (cpumask_empty(nodemask))
7544
7545 *nodemask = node_to_cpumask(i);
7546 cpus_and(*nodemask, *nodemask, *cpu_map);
7547 if (cpus_empty(*nodemask))
7548 continue; 7646 continue;
7549 7647
7550 init_sched_build_groups(nodemask, cpu_map, 7648 init_sched_build_groups(nodemask, cpu_map,
@@ -7555,8 +7653,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7555#ifdef CONFIG_NUMA 7653#ifdef CONFIG_NUMA
7556 /* Set up node groups */ 7654 /* Set up node groups */
7557 if (sd_allnodes) { 7655 if (sd_allnodes) {
7558 SCHED_CPUMASK_VAR(send_covered, allmasks);
7559
7560 init_sched_build_groups(cpu_map, cpu_map, 7656 init_sched_build_groups(cpu_map, cpu_map,
7561 &cpu_to_allnodes_group, 7657 &cpu_to_allnodes_group,
7562 send_covered, tmpmask); 7658 send_covered, tmpmask);
@@ -7565,58 +7661,53 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7565 for (i = 0; i < nr_node_ids; i++) { 7661 for (i = 0; i < nr_node_ids; i++) {
7566 /* Set up node groups */ 7662 /* Set up node groups */
7567 struct sched_group *sg, *prev; 7663 struct sched_group *sg, *prev;
7568 SCHED_CPUMASK_VAR(nodemask, allmasks);
7569 SCHED_CPUMASK_VAR(domainspan, allmasks);
7570 SCHED_CPUMASK_VAR(covered, allmasks);
7571 int j; 7664 int j;
7572 7665
7573 *nodemask = node_to_cpumask(i); 7666 cpumask_clear(covered);
7574 cpus_clear(*covered); 7667 cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
7575 7668 if (cpumask_empty(nodemask)) {
7576 cpus_and(*nodemask, *nodemask, *cpu_map);
7577 if (cpus_empty(*nodemask)) {
7578 sched_group_nodes[i] = NULL; 7669 sched_group_nodes[i] = NULL;
7579 continue; 7670 continue;
7580 } 7671 }
7581 7672
7582 sched_domain_node_span(i, domainspan); 7673 sched_domain_node_span(i, domainspan);
7583 cpus_and(*domainspan, *domainspan, *cpu_map); 7674 cpumask_and(domainspan, domainspan, cpu_map);
7584 7675
7585 sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); 7676 sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
7677 GFP_KERNEL, i);
7586 if (!sg) { 7678 if (!sg) {
7587 printk(KERN_WARNING "Can not alloc domain group for " 7679 printk(KERN_WARNING "Can not alloc domain group for "
7588 "node %d\n", i); 7680 "node %d\n", i);
7589 goto error; 7681 goto error;
7590 } 7682 }
7591 sched_group_nodes[i] = sg; 7683 sched_group_nodes[i] = sg;
7592 for_each_cpu_mask_nr(j, *nodemask) { 7684 for_each_cpu(j, nodemask) {
7593 struct sched_domain *sd; 7685 struct sched_domain *sd;
7594 7686
7595 sd = &per_cpu(node_domains, j); 7687 sd = &per_cpu(node_domains, j);
7596 sd->groups = sg; 7688 sd->groups = sg;
7597 } 7689 }
7598 sg->__cpu_power = 0; 7690 sg->__cpu_power = 0;
7599 sg->cpumask = *nodemask; 7691 cpumask_copy(sched_group_cpus(sg), nodemask);
7600 sg->next = sg; 7692 sg->next = sg;
7601 cpus_or(*covered, *covered, *nodemask); 7693 cpumask_or(covered, covered, nodemask);
7602 prev = sg; 7694 prev = sg;
7603 7695
7604 for (j = 0; j < nr_node_ids; j++) { 7696 for (j = 0; j < nr_node_ids; j++) {
7605 SCHED_CPUMASK_VAR(notcovered, allmasks);
7606 int n = (i + j) % nr_node_ids; 7697 int n = (i + j) % nr_node_ids;
7607 node_to_cpumask_ptr(pnodemask, n);
7608 7698
7609 cpus_complement(*notcovered, *covered); 7699 cpumask_complement(notcovered, covered);
7610 cpus_and(*tmpmask, *notcovered, *cpu_map); 7700 cpumask_and(tmpmask, notcovered, cpu_map);
7611 cpus_and(*tmpmask, *tmpmask, *domainspan); 7701 cpumask_and(tmpmask, tmpmask, domainspan);
7612 if (cpus_empty(*tmpmask)) 7702 if (cpumask_empty(tmpmask))
7613 break; 7703 break;
7614 7704
7615 cpus_and(*tmpmask, *tmpmask, *pnodemask); 7705 cpumask_and(tmpmask, tmpmask, cpumask_of_node(n));
7616 if (cpus_empty(*tmpmask)) 7706 if (cpumask_empty(tmpmask))
7617 continue; 7707 continue;
7618 7708
7619 sg = kmalloc_node(sizeof(struct sched_group), 7709 sg = kmalloc_node(sizeof(struct sched_group) +
7710 cpumask_size(),
7620 GFP_KERNEL, i); 7711 GFP_KERNEL, i);
7621 if (!sg) { 7712 if (!sg) {
7622 printk(KERN_WARNING 7713 printk(KERN_WARNING
@@ -7624,9 +7715,9 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7624 goto error; 7715 goto error;
7625 } 7716 }
7626 sg->__cpu_power = 0; 7717 sg->__cpu_power = 0;
7627 sg->cpumask = *tmpmask; 7718 cpumask_copy(sched_group_cpus(sg), tmpmask);
7628 sg->next = prev->next; 7719 sg->next = prev->next;
7629 cpus_or(*covered, *covered, *tmpmask); 7720 cpumask_or(covered, covered, tmpmask);
7630 prev->next = sg; 7721 prev->next = sg;
7631 prev = sg; 7722 prev = sg;
7632 } 7723 }
@@ -7635,22 +7726,22 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7635 7726
7636 /* Calculate CPU power for physical packages and nodes */ 7727 /* Calculate CPU power for physical packages and nodes */
7637#ifdef CONFIG_SCHED_SMT 7728#ifdef CONFIG_SCHED_SMT
7638 for_each_cpu_mask_nr(i, *cpu_map) { 7729 for_each_cpu(i, cpu_map) {
7639 struct sched_domain *sd = &per_cpu(cpu_domains, i); 7730 struct sched_domain *sd = &per_cpu(cpu_domains, i).sd;
7640 7731
7641 init_sched_groups_power(i, sd); 7732 init_sched_groups_power(i, sd);
7642 } 7733 }
7643#endif 7734#endif
7644#ifdef CONFIG_SCHED_MC 7735#ifdef CONFIG_SCHED_MC
7645 for_each_cpu_mask_nr(i, *cpu_map) { 7736 for_each_cpu(i, cpu_map) {
7646 struct sched_domain *sd = &per_cpu(core_domains, i); 7737 struct sched_domain *sd = &per_cpu(core_domains, i).sd;
7647 7738
7648 init_sched_groups_power(i, sd); 7739 init_sched_groups_power(i, sd);
7649 } 7740 }
7650#endif 7741#endif
7651 7742
7652 for_each_cpu_mask_nr(i, *cpu_map) { 7743 for_each_cpu(i, cpu_map) {
7653 struct sched_domain *sd = &per_cpu(phys_domains, i); 7744 struct sched_domain *sd = &per_cpu(phys_domains, i).sd;
7654 7745
7655 init_sched_groups_power(i, sd); 7746 init_sched_groups_power(i, sd);
7656 } 7747 }
@@ -7662,56 +7753,87 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7662 if (sd_allnodes) { 7753 if (sd_allnodes) {
7663 struct sched_group *sg; 7754 struct sched_group *sg;
7664 7755
7665 cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, 7756 cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
7666 tmpmask); 7757 tmpmask);
7667 init_numa_sched_groups_power(sg); 7758 init_numa_sched_groups_power(sg);
7668 } 7759 }
7669#endif 7760#endif
7670 7761
7671 /* Attach the domains */ 7762 /* Attach the domains */
7672 for_each_cpu_mask_nr(i, *cpu_map) { 7763 for_each_cpu(i, cpu_map) {
7673 struct sched_domain *sd; 7764 struct sched_domain *sd;
7674#ifdef CONFIG_SCHED_SMT 7765#ifdef CONFIG_SCHED_SMT
7675 sd = &per_cpu(cpu_domains, i); 7766 sd = &per_cpu(cpu_domains, i).sd;
7676#elif defined(CONFIG_SCHED_MC) 7767#elif defined(CONFIG_SCHED_MC)
7677 sd = &per_cpu(core_domains, i); 7768 sd = &per_cpu(core_domains, i).sd;
7678#else 7769#else
7679 sd = &per_cpu(phys_domains, i); 7770 sd = &per_cpu(phys_domains, i).sd;
7680#endif 7771#endif
7681 cpu_attach_domain(sd, rd, i); 7772 cpu_attach_domain(sd, rd, i);
7682 } 7773 }
7683 7774
7684 SCHED_CPUMASK_FREE((void *)allmasks); 7775 err = 0;
7685 return 0; 7776
7777free_tmpmask:
7778 free_cpumask_var(tmpmask);
7779free_send_covered:
7780 free_cpumask_var(send_covered);
7781free_this_core_map:
7782 free_cpumask_var(this_core_map);
7783free_this_sibling_map:
7784 free_cpumask_var(this_sibling_map);
7785free_nodemask:
7786 free_cpumask_var(nodemask);
7787free_notcovered:
7788#ifdef CONFIG_NUMA
7789 free_cpumask_var(notcovered);
7790free_covered:
7791 free_cpumask_var(covered);
7792free_domainspan:
7793 free_cpumask_var(domainspan);
7794out:
7795#endif
7796 return err;
7797
7798free_sched_groups:
7799#ifdef CONFIG_NUMA
7800 kfree(sched_group_nodes);
7801#endif
7802 goto free_tmpmask;
7686 7803
7687#ifdef CONFIG_NUMA 7804#ifdef CONFIG_NUMA
7688error: 7805error:
7689 free_sched_groups(cpu_map, tmpmask); 7806 free_sched_groups(cpu_map, tmpmask);
7690 SCHED_CPUMASK_FREE((void *)allmasks); 7807 free_rootdomain(rd);
7691 kfree(rd); 7808 goto free_tmpmask;
7692 return -ENOMEM;
7693#endif 7809#endif
7694} 7810}
7695 7811
7696static int build_sched_domains(const cpumask_t *cpu_map) 7812static int build_sched_domains(const struct cpumask *cpu_map)
7697{ 7813{
7698 return __build_sched_domains(cpu_map, NULL); 7814 return __build_sched_domains(cpu_map, NULL);
7699} 7815}
7700 7816
7701static cpumask_t *doms_cur; /* current sched domains */ 7817static struct cpumask *doms_cur; /* current sched domains */
7702static int ndoms_cur; /* number of sched domains in 'doms_cur' */ 7818static int ndoms_cur; /* number of sched domains in 'doms_cur' */
7703static struct sched_domain_attr *dattr_cur; 7819static struct sched_domain_attr *dattr_cur;
7704 /* attribues of custom domains in 'doms_cur' */ 7820 /* attribues of custom domains in 'doms_cur' */
7705 7821
7706/* 7822/*
7707 * Special case: If a kmalloc of a doms_cur partition (array of 7823 * Special case: If a kmalloc of a doms_cur partition (array of
7708 * cpumask_t) fails, then fallback to a single sched domain, 7824 * cpumask) fails, then fallback to a single sched domain,
7709 * as determined by the single cpumask_t fallback_doms. 7825 * as determined by the single cpumask fallback_doms.
7710 */ 7826 */
7711static cpumask_t fallback_doms; 7827static cpumask_var_t fallback_doms;
7712 7828
7713void __attribute__((weak)) arch_update_cpu_topology(void) 7829/*
7830 * arch_update_cpu_topology lets virtualized architectures update the
7831 * cpu core maps. It is supposed to return 1 if the topology changed
7832 * or 0 if it stayed the same.
7833 */
7834int __attribute__((weak)) arch_update_cpu_topology(void)
7714{ 7835{
7836 return 0;
7715} 7837}
7716 7838
7717/* 7839/*
@@ -7719,16 +7841,16 @@ void __attribute__((weak)) arch_update_cpu_topology(void)
7719 * For now this just excludes isolated cpus, but could be used to 7841 * For now this just excludes isolated cpus, but could be used to
7720 * exclude other special cases in the future. 7842 * exclude other special cases in the future.
7721 */ 7843 */
7722static int arch_init_sched_domains(const cpumask_t *cpu_map) 7844static int arch_init_sched_domains(const struct cpumask *cpu_map)
7723{ 7845{
7724 int err; 7846 int err;
7725 7847
7726 arch_update_cpu_topology(); 7848 arch_update_cpu_topology();
7727 ndoms_cur = 1; 7849 ndoms_cur = 1;
7728 doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); 7850 doms_cur = kmalloc(cpumask_size(), GFP_KERNEL);
7729 if (!doms_cur) 7851 if (!doms_cur)
7730 doms_cur = &fallback_doms; 7852 doms_cur = fallback_doms;
7731 cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); 7853 cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map);
7732 dattr_cur = NULL; 7854 dattr_cur = NULL;
7733 err = build_sched_domains(doms_cur); 7855 err = build_sched_domains(doms_cur);
7734 register_sched_domain_sysctl(); 7856 register_sched_domain_sysctl();
@@ -7736,8 +7858,8 @@ static int arch_init_sched_domains(const cpumask_t *cpu_map)
7736 return err; 7858 return err;
7737} 7859}
7738 7860
7739static void arch_destroy_sched_domains(const cpumask_t *cpu_map, 7861static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
7740 cpumask_t *tmpmask) 7862 struct cpumask *tmpmask)
7741{ 7863{
7742 free_sched_groups(cpu_map, tmpmask); 7864 free_sched_groups(cpu_map, tmpmask);
7743} 7865}
@@ -7746,17 +7868,16 @@ static void arch_destroy_sched_domains(const cpumask_t *cpu_map,
7746 * Detach sched domains from a group of cpus specified in cpu_map 7868 * Detach sched domains from a group of cpus specified in cpu_map
7747 * These cpus will now be attached to the NULL domain 7869 * These cpus will now be attached to the NULL domain
7748 */ 7870 */
7749static void detach_destroy_domains(const cpumask_t *cpu_map) 7871static void detach_destroy_domains(const struct cpumask *cpu_map)
7750{ 7872{
7751 cpumask_t tmpmask; 7873 /* Save because hotplug lock held. */
7874 static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
7752 int i; 7875 int i;
7753 7876
7754 unregister_sched_domain_sysctl(); 7877 for_each_cpu(i, cpu_map)
7755
7756 for_each_cpu_mask_nr(i, *cpu_map)
7757 cpu_attach_domain(NULL, &def_root_domain, i); 7878 cpu_attach_domain(NULL, &def_root_domain, i);
7758 synchronize_sched(); 7879 synchronize_sched();
7759 arch_destroy_sched_domains(cpu_map, &tmpmask); 7880 arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
7760} 7881}
7761 7882
7762/* handle null as "default" */ 7883/* handle null as "default" */
@@ -7781,7 +7902,7 @@ static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7781 * doms_new[] to the current sched domain partitioning, doms_cur[]. 7902 * doms_new[] to the current sched domain partitioning, doms_cur[].
7782 * It destroys each deleted domain and builds each new domain. 7903 * It destroys each deleted domain and builds each new domain.
7783 * 7904 *
7784 * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. 7905 * 'doms_new' is an array of cpumask's of length 'ndoms_new'.
7785 * The masks don't intersect (don't overlap.) We should setup one 7906 * The masks don't intersect (don't overlap.) We should setup one
7786 * sched domain for each mask. CPUs not in any of the cpumasks will 7907 * sched domain for each mask. CPUs not in any of the cpumasks will
7787 * not be load balanced. If the same cpumask appears both in the 7908 * not be load balanced. If the same cpumask appears both in the
@@ -7795,28 +7916,33 @@ static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
7795 * the single partition 'fallback_doms', it also forces the domains 7916 * the single partition 'fallback_doms', it also forces the domains
7796 * to be rebuilt. 7917 * to be rebuilt.
7797 * 7918 *
7798 * If doms_new == NULL it will be replaced with cpu_online_map. 7919 * If doms_new == NULL it will be replaced with cpu_online_mask.
7799 * ndoms_new == 0 is a special case for destroying existing domains, 7920 * ndoms_new == 0 is a special case for destroying existing domains,
7800 * and it will not create the default domain. 7921 * and it will not create the default domain.
7801 * 7922 *
7802 * Call with hotplug lock held 7923 * Call with hotplug lock held
7803 */ 7924 */
7804void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, 7925/* FIXME: Change to struct cpumask *doms_new[] */
7926void partition_sched_domains(int ndoms_new, struct cpumask *doms_new,
7805 struct sched_domain_attr *dattr_new) 7927 struct sched_domain_attr *dattr_new)
7806{ 7928{
7807 int i, j, n; 7929 int i, j, n;
7930 int new_topology;
7808 7931
7809 mutex_lock(&sched_domains_mutex); 7932 mutex_lock(&sched_domains_mutex);
7810 7933
7811 /* always unregister in case we don't destroy any domains */ 7934 /* always unregister in case we don't destroy any domains */
7812 unregister_sched_domain_sysctl(); 7935 unregister_sched_domain_sysctl();
7813 7936
7937 /* Let architecture update cpu core mappings. */
7938 new_topology = arch_update_cpu_topology();
7939
7814 n = doms_new ? ndoms_new : 0; 7940 n = doms_new ? ndoms_new : 0;
7815 7941
7816 /* Destroy deleted domains */ 7942 /* Destroy deleted domains */
7817 for (i = 0; i < ndoms_cur; i++) { 7943 for (i = 0; i < ndoms_cur; i++) {
7818 for (j = 0; j < n; j++) { 7944 for (j = 0; j < n && !new_topology; j++) {
7819 if (cpus_equal(doms_cur[i], doms_new[j]) 7945 if (cpumask_equal(&doms_cur[i], &doms_new[j])
7820 && dattrs_equal(dattr_cur, i, dattr_new, j)) 7946 && dattrs_equal(dattr_cur, i, dattr_new, j))
7821 goto match1; 7947 goto match1;
7822 } 7948 }
@@ -7828,15 +7954,15 @@ match1:
7828 7954
7829 if (doms_new == NULL) { 7955 if (doms_new == NULL) {
7830 ndoms_cur = 0; 7956 ndoms_cur = 0;
7831 doms_new = &fallback_doms; 7957 doms_new = fallback_doms;
7832 cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); 7958 cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map);
7833 dattr_new = NULL; 7959 WARN_ON_ONCE(dattr_new);
7834 } 7960 }
7835 7961
7836 /* Build new domains */ 7962 /* Build new domains */
7837 for (i = 0; i < ndoms_new; i++) { 7963 for (i = 0; i < ndoms_new; i++) {
7838 for (j = 0; j < ndoms_cur; j++) { 7964 for (j = 0; j < ndoms_cur && !new_topology; j++) {
7839 if (cpus_equal(doms_new[i], doms_cur[j]) 7965 if (cpumask_equal(&doms_new[i], &doms_cur[j])
7840 && dattrs_equal(dattr_new, i, dattr_cur, j)) 7966 && dattrs_equal(dattr_new, i, dattr_cur, j))
7841 goto match2; 7967 goto match2;
7842 } 7968 }
@@ -7848,7 +7974,7 @@ match2:
7848 } 7974 }
7849 7975
7850 /* Remember the new sched domains */ 7976 /* Remember the new sched domains */
7851 if (doms_cur != &fallback_doms) 7977 if (doms_cur != fallback_doms)
7852 kfree(doms_cur); 7978 kfree(doms_cur);
7853 kfree(dattr_cur); /* kfree(NULL) is safe */ 7979 kfree(dattr_cur); /* kfree(NULL) is safe */
7854 doms_cur = doms_new; 7980 doms_cur = doms_new;
@@ -7877,14 +8003,25 @@ int arch_reinit_sched_domains(void)
7877static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) 8003static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt)
7878{ 8004{
7879 int ret; 8005 int ret;
8006 unsigned int level = 0;
8007
8008 if (sscanf(buf, "%u", &level) != 1)
8009 return -EINVAL;
8010
8011 /*
8012 * level is always be positive so don't check for
8013 * level < POWERSAVINGS_BALANCE_NONE which is 0
8014 * What happens on 0 or 1 byte write,
8015 * need to check for count as well?
8016 */
7880 8017
7881 if (buf[0] != '0' && buf[0] != '1') 8018 if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS)
7882 return -EINVAL; 8019 return -EINVAL;
7883 8020
7884 if (smt) 8021 if (smt)
7885 sched_smt_power_savings = (buf[0] == '1'); 8022 sched_smt_power_savings = level;
7886 else 8023 else
7887 sched_mc_power_savings = (buf[0] == '1'); 8024 sched_mc_power_savings = level;
7888 8025
7889 ret = arch_reinit_sched_domains(); 8026 ret = arch_reinit_sched_domains();
7890 8027
@@ -7988,7 +8125,9 @@ static int update_runtime(struct notifier_block *nfb,
7988 8125
7989void __init sched_init_smp(void) 8126void __init sched_init_smp(void)
7990{ 8127{
7991 cpumask_t non_isolated_cpus; 8128 cpumask_var_t non_isolated_cpus;
8129
8130 alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7992 8131
7993#if defined(CONFIG_NUMA) 8132#if defined(CONFIG_NUMA)
7994 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), 8133 sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
@@ -7997,10 +8136,10 @@ void __init sched_init_smp(void)
7997#endif 8136#endif
7998 get_online_cpus(); 8137 get_online_cpus();
7999 mutex_lock(&sched_domains_mutex); 8138 mutex_lock(&sched_domains_mutex);
8000 arch_init_sched_domains(&cpu_online_map); 8139 arch_init_sched_domains(cpu_online_mask);
8001 cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); 8140 cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
8002 if (cpus_empty(non_isolated_cpus)) 8141 if (cpumask_empty(non_isolated_cpus))
8003 cpu_set(smp_processor_id(), non_isolated_cpus); 8142 cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
8004 mutex_unlock(&sched_domains_mutex); 8143 mutex_unlock(&sched_domains_mutex);
8005 put_online_cpus(); 8144 put_online_cpus();
8006 8145
@@ -8015,9 +8154,13 @@ void __init sched_init_smp(void)
8015 init_hrtick(); 8154 init_hrtick();
8016 8155
8017 /* Move init over to a non-isolated CPU */ 8156 /* Move init over to a non-isolated CPU */
8018 if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) 8157 if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
8019 BUG(); 8158 BUG();
8020 sched_init_granularity(); 8159 sched_init_granularity();
8160 free_cpumask_var(non_isolated_cpus);
8161
8162 alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
8163 init_sched_rt_class();
8021} 8164}
8022#else 8165#else
8023void __init sched_init_smp(void) 8166void __init sched_init_smp(void)
@@ -8332,6 +8475,15 @@ void __init sched_init(void)
8332 */ 8475 */
8333 current->sched_class = &fair_sched_class; 8476 current->sched_class = &fair_sched_class;
8334 8477
8478 /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */
8479 alloc_bootmem_cpumask_var(&nohz_cpu_mask);
8480#ifdef CONFIG_SMP
8481#ifdef CONFIG_NO_HZ
8482 alloc_bootmem_cpumask_var(&nohz.cpu_mask);
8483#endif
8484 alloc_bootmem_cpumask_var(&cpu_isolated_map);
8485#endif /* SMP */
8486
8335 scheduler_running = 1; 8487 scheduler_running = 1;
8336} 8488}
8337 8489
@@ -8490,7 +8642,7 @@ static
8490int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) 8642int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8491{ 8643{
8492 struct cfs_rq *cfs_rq; 8644 struct cfs_rq *cfs_rq;
8493 struct sched_entity *se, *parent_se; 8645 struct sched_entity *se;
8494 struct rq *rq; 8646 struct rq *rq;
8495 int i; 8647 int i;
8496 8648
@@ -8506,18 +8658,17 @@ int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8506 for_each_possible_cpu(i) { 8658 for_each_possible_cpu(i) {
8507 rq = cpu_rq(i); 8659 rq = cpu_rq(i);
8508 8660
8509 cfs_rq = kmalloc_node(sizeof(struct cfs_rq), 8661 cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
8510 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8662 GFP_KERNEL, cpu_to_node(i));
8511 if (!cfs_rq) 8663 if (!cfs_rq)
8512 goto err; 8664 goto err;
8513 8665
8514 se = kmalloc_node(sizeof(struct sched_entity), 8666 se = kzalloc_node(sizeof(struct sched_entity),
8515 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8667 GFP_KERNEL, cpu_to_node(i));
8516 if (!se) 8668 if (!se)
8517 goto err; 8669 goto err;
8518 8670
8519 parent_se = parent ? parent->se[i] : NULL; 8671 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]);
8520 init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se);
8521 } 8672 }
8522 8673
8523 return 1; 8674 return 1;
@@ -8578,7 +8729,7 @@ static
8578int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) 8729int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8579{ 8730{
8580 struct rt_rq *rt_rq; 8731 struct rt_rq *rt_rq;
8581 struct sched_rt_entity *rt_se, *parent_se; 8732 struct sched_rt_entity *rt_se;
8582 struct rq *rq; 8733 struct rq *rq;
8583 int i; 8734 int i;
8584 8735
@@ -8595,18 +8746,17 @@ int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8595 for_each_possible_cpu(i) { 8746 for_each_possible_cpu(i) {
8596 rq = cpu_rq(i); 8747 rq = cpu_rq(i);
8597 8748
8598 rt_rq = kmalloc_node(sizeof(struct rt_rq), 8749 rt_rq = kzalloc_node(sizeof(struct rt_rq),
8599 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8750 GFP_KERNEL, cpu_to_node(i));
8600 if (!rt_rq) 8751 if (!rt_rq)
8601 goto err; 8752 goto err;
8602 8753
8603 rt_se = kmalloc_node(sizeof(struct sched_rt_entity), 8754 rt_se = kzalloc_node(sizeof(struct sched_rt_entity),
8604 GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); 8755 GFP_KERNEL, cpu_to_node(i));
8605 if (!rt_se) 8756 if (!rt_se)
8606 goto err; 8757 goto err;
8607 8758
8608 parent_se = parent ? parent->rt_se[i] : NULL; 8759 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]);
8609 init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se);
8610 } 8760 }
8611 8761
8612 return 1; 8762 return 1;
@@ -9249,11 +9399,12 @@ struct cgroup_subsys cpu_cgroup_subsys = {
9249 * (balbir@in.ibm.com). 9399 * (balbir@in.ibm.com).
9250 */ 9400 */
9251 9401
9252/* track cpu usage of a group of tasks */ 9402/* track cpu usage of a group of tasks and its child groups */
9253struct cpuacct { 9403struct cpuacct {
9254 struct cgroup_subsys_state css; 9404 struct cgroup_subsys_state css;
9255 /* cpuusage holds pointer to a u64-type object on every cpu */ 9405 /* cpuusage holds pointer to a u64-type object on every cpu */
9256 u64 *cpuusage; 9406 u64 *cpuusage;
9407 struct cpuacct *parent;
9257}; 9408};
9258 9409
9259struct cgroup_subsys cpuacct_subsys; 9410struct cgroup_subsys cpuacct_subsys;
@@ -9287,6 +9438,9 @@ static struct cgroup_subsys_state *cpuacct_create(
9287 return ERR_PTR(-ENOMEM); 9438 return ERR_PTR(-ENOMEM);
9288 } 9439 }
9289 9440
9441 if (cgrp->parent)
9442 ca->parent = cgroup_ca(cgrp->parent);
9443
9290 return &ca->css; 9444 return &ca->css;
9291} 9445}
9292 9446
@@ -9300,6 +9454,41 @@ cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
9300 kfree(ca); 9454 kfree(ca);
9301} 9455}
9302 9456
9457static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
9458{
9459 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu);
9460 u64 data;
9461
9462#ifndef CONFIG_64BIT
9463 /*
9464 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
9465 */
9466 spin_lock_irq(&cpu_rq(cpu)->lock);
9467 data = *cpuusage;
9468 spin_unlock_irq(&cpu_rq(cpu)->lock);
9469#else
9470 data = *cpuusage;
9471#endif
9472
9473 return data;
9474}
9475
9476static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
9477{
9478 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu);
9479
9480#ifndef CONFIG_64BIT
9481 /*
9482 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
9483 */
9484 spin_lock_irq(&cpu_rq(cpu)->lock);
9485 *cpuusage = val;
9486 spin_unlock_irq(&cpu_rq(cpu)->lock);
9487#else
9488 *cpuusage = val;
9489#endif
9490}
9491
9303/* return total cpu usage (in nanoseconds) of a group */ 9492/* return total cpu usage (in nanoseconds) of a group */
9304static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) 9493static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
9305{ 9494{
@@ -9307,17 +9496,8 @@ static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft)
9307 u64 totalcpuusage = 0; 9496 u64 totalcpuusage = 0;
9308 int i; 9497 int i;
9309 9498
9310 for_each_possible_cpu(i) { 9499 for_each_present_cpu(i)
9311 u64 *cpuusage = percpu_ptr(ca->cpuusage, i); 9500 totalcpuusage += cpuacct_cpuusage_read(ca, i);
9312
9313 /*
9314 * Take rq->lock to make 64-bit addition safe on 32-bit
9315 * platforms.
9316 */
9317 spin_lock_irq(&cpu_rq(i)->lock);
9318 totalcpuusage += *cpuusage;
9319 spin_unlock_irq(&cpu_rq(i)->lock);
9320 }
9321 9501
9322 return totalcpuusage; 9502 return totalcpuusage;
9323} 9503}
@@ -9334,23 +9514,39 @@ static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype,
9334 goto out; 9514 goto out;
9335 } 9515 }
9336 9516
9337 for_each_possible_cpu(i) { 9517 for_each_present_cpu(i)
9338 u64 *cpuusage = percpu_ptr(ca->cpuusage, i); 9518 cpuacct_cpuusage_write(ca, i, 0);
9339 9519
9340 spin_lock_irq(&cpu_rq(i)->lock);
9341 *cpuusage = 0;
9342 spin_unlock_irq(&cpu_rq(i)->lock);
9343 }
9344out: 9520out:
9345 return err; 9521 return err;
9346} 9522}
9347 9523
9524static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft,
9525 struct seq_file *m)
9526{
9527 struct cpuacct *ca = cgroup_ca(cgroup);
9528 u64 percpu;
9529 int i;
9530
9531 for_each_present_cpu(i) {
9532 percpu = cpuacct_cpuusage_read(ca, i);
9533 seq_printf(m, "%llu ", (unsigned long long) percpu);
9534 }
9535 seq_printf(m, "\n");
9536 return 0;
9537}
9538
9348static struct cftype files[] = { 9539static struct cftype files[] = {
9349 { 9540 {
9350 .name = "usage", 9541 .name = "usage",
9351 .read_u64 = cpuusage_read, 9542 .read_u64 = cpuusage_read,
9352 .write_u64 = cpuusage_write, 9543 .write_u64 = cpuusage_write,
9353 }, 9544 },
9545 {
9546 .name = "usage_percpu",
9547 .read_seq_string = cpuacct_percpu_seq_read,
9548 },
9549
9354}; 9550};
9355 9551
9356static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) 9552static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
@@ -9366,14 +9562,16 @@ static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
9366static void cpuacct_charge(struct task_struct *tsk, u64 cputime) 9562static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9367{ 9563{
9368 struct cpuacct *ca; 9564 struct cpuacct *ca;
9565 int cpu;
9369 9566
9370 if (!cpuacct_subsys.active) 9567 if (!cpuacct_subsys.active)
9371 return; 9568 return;
9372 9569
9570 cpu = task_cpu(tsk);
9373 ca = task_ca(tsk); 9571 ca = task_ca(tsk);
9374 if (ca) {
9375 u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk));
9376 9572
9573 for (; ca; ca = ca->parent) {
9574 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu);
9377 *cpuusage += cputime; 9575 *cpuusage += cputime;
9378 } 9576 }
9379} 9577}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-25 02:09:27 -0500