diff options
Diffstat (limited to 'kernel/sched.c')
-rw-r--r-- | kernel/sched.c | 43 |
1 files changed, 22 insertions, 21 deletions
diff --git a/kernel/sched.c b/kernel/sched.c index 5757e03cfac0..0ad7c28b7236 100644 --- a/kernel/sched.c +++ b/kernel/sched.c | |||
@@ -3818,19 +3818,23 @@ find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, | |||
3818 | */ | 3818 | */ |
3819 | #define MAX_PINNED_INTERVAL 512 | 3819 | #define MAX_PINNED_INTERVAL 512 |
3820 | 3820 | ||
3821 | /* Working cpumask for load_balance and load_balance_newidle. */ | ||
3822 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | ||
3823 | |||
3821 | /* | 3824 | /* |
3822 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | 3825 | * Check this_cpu to ensure it is balanced within domain. Attempt to move |
3823 | * tasks if there is an imbalance. | 3826 | * tasks if there is an imbalance. |
3824 | */ | 3827 | */ |
3825 | static int load_balance(int this_cpu, struct rq *this_rq, | 3828 | static int load_balance(int this_cpu, struct rq *this_rq, |
3826 | struct sched_domain *sd, enum cpu_idle_type idle, | 3829 | struct sched_domain *sd, enum cpu_idle_type idle, |
3827 | int *balance, struct cpumask *cpus) | 3830 | int *balance) |
3828 | { | 3831 | { |
3829 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; | 3832 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
3830 | struct sched_group *group; | 3833 | struct sched_group *group; |
3831 | unsigned long imbalance; | 3834 | unsigned long imbalance; |
3832 | struct rq *busiest; | 3835 | struct rq *busiest; |
3833 | unsigned long flags; | 3836 | unsigned long flags; |
3837 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); | ||
3834 | 3838 | ||
3835 | cpumask_setall(cpus); | 3839 | cpumask_setall(cpus); |
3836 | 3840 | ||
@@ -3985,8 +3989,7 @@ out: | |||
3985 | * this_rq is locked. | 3989 | * this_rq is locked. |
3986 | */ | 3990 | */ |
3987 | static int | 3991 | static int |
3988 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, | 3992 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
3989 | struct cpumask *cpus) | ||
3990 | { | 3993 | { |
3991 | struct sched_group *group; | 3994 | struct sched_group *group; |
3992 | struct rq *busiest = NULL; | 3995 | struct rq *busiest = NULL; |
@@ -3994,6 +3997,7 @@ load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, | |||
3994 | int ld_moved = 0; | 3997 | int ld_moved = 0; |
3995 | int sd_idle = 0; | 3998 | int sd_idle = 0; |
3996 | int all_pinned = 0; | 3999 | int all_pinned = 0; |
4000 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); | ||
3997 | 4001 | ||
3998 | cpumask_setall(cpus); | 4002 | cpumask_setall(cpus); |
3999 | 4003 | ||
@@ -4134,10 +4138,6 @@ static void idle_balance(int this_cpu, struct rq *this_rq) | |||
4134 | struct sched_domain *sd; | 4138 | struct sched_domain *sd; |
4135 | int pulled_task = 0; | 4139 | int pulled_task = 0; |
4136 | unsigned long next_balance = jiffies + HZ; | 4140 | unsigned long next_balance = jiffies + HZ; |
4137 | cpumask_var_t tmpmask; | ||
4138 | |||
4139 | if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC)) | ||
4140 | return; | ||
4141 | 4141 | ||
4142 | for_each_domain(this_cpu, sd) { | 4142 | for_each_domain(this_cpu, sd) { |
4143 | unsigned long interval; | 4143 | unsigned long interval; |
@@ -4148,7 +4148,7 @@ static void idle_balance(int this_cpu, struct rq *this_rq) | |||
4148 | if (sd->flags & SD_BALANCE_NEWIDLE) | 4148 | if (sd->flags & SD_BALANCE_NEWIDLE) |
4149 | /* If we've pulled tasks over stop searching: */ | 4149 | /* If we've pulled tasks over stop searching: */ |
4150 | pulled_task = load_balance_newidle(this_cpu, this_rq, | 4150 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
4151 | sd, tmpmask); | 4151 | sd); |
4152 | 4152 | ||
4153 | interval = msecs_to_jiffies(sd->balance_interval); | 4153 | interval = msecs_to_jiffies(sd->balance_interval); |
4154 | if (time_after(next_balance, sd->last_balance + interval)) | 4154 | if (time_after(next_balance, sd->last_balance + interval)) |
@@ -4163,7 +4163,6 @@ static void idle_balance(int this_cpu, struct rq *this_rq) | |||
4163 | */ | 4163 | */ |
4164 | this_rq->next_balance = next_balance; | 4164 | this_rq->next_balance = next_balance; |
4165 | } | 4165 | } |
4166 | free_cpumask_var(tmpmask); | ||
4167 | } | 4166 | } |
4168 | 4167 | ||
4169 | /* | 4168 | /* |
@@ -4313,11 +4312,6 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle) | |||
4313 | unsigned long next_balance = jiffies + 60*HZ; | 4312 | unsigned long next_balance = jiffies + 60*HZ; |
4314 | int update_next_balance = 0; | 4313 | int update_next_balance = 0; |
4315 | int need_serialize; | 4314 | int need_serialize; |
4316 | cpumask_var_t tmp; | ||
4317 | |||
4318 | /* Fails alloc? Rebalancing probably not a priority right now. */ | ||
4319 | if (!alloc_cpumask_var(&tmp, GFP_ATOMIC)) | ||
4320 | return; | ||
4321 | 4315 | ||
4322 | for_each_domain(cpu, sd) { | 4316 | for_each_domain(cpu, sd) { |
4323 | if (!(sd->flags & SD_LOAD_BALANCE)) | 4317 | if (!(sd->flags & SD_LOAD_BALANCE)) |
@@ -4342,7 +4336,7 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle) | |||
4342 | } | 4336 | } |
4343 | 4337 | ||
4344 | if (time_after_eq(jiffies, sd->last_balance + interval)) { | 4338 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
4345 | if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { | 4339 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
4346 | /* | 4340 | /* |
4347 | * We've pulled tasks over so either we're no | 4341 | * We've pulled tasks over so either we're no |
4348 | * longer idle, or one of our SMT siblings is | 4342 | * longer idle, or one of our SMT siblings is |
@@ -4376,8 +4370,6 @@ out: | |||
4376 | */ | 4370 | */ |
4377 | if (likely(update_next_balance)) | 4371 | if (likely(update_next_balance)) |
4378 | rq->next_balance = next_balance; | 4372 | rq->next_balance = next_balance; |
4379 | |||
4380 | free_cpumask_var(tmp); | ||
4381 | } | 4373 | } |
4382 | 4374 | ||
4383 | /* | 4375 | /* |
@@ -7648,7 +7640,7 @@ cpu_to_core_group(int cpu, const struct cpumask *cpu_map, | |||
7648 | { | 7640 | { |
7649 | int group; | 7641 | int group; |
7650 | 7642 | ||
7651 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); | 7643 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
7652 | group = cpumask_first(mask); | 7644 | group = cpumask_first(mask); |
7653 | if (sg) | 7645 | if (sg) |
7654 | *sg = &per_cpu(sched_group_core, group).sg; | 7646 | *sg = &per_cpu(sched_group_core, group).sg; |
@@ -7677,7 +7669,7 @@ cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, | |||
7677 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | 7669 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
7678 | group = cpumask_first(mask); | 7670 | group = cpumask_first(mask); |
7679 | #elif defined(CONFIG_SCHED_SMT) | 7671 | #elif defined(CONFIG_SCHED_SMT) |
7680 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); | 7672 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
7681 | group = cpumask_first(mask); | 7673 | group = cpumask_first(mask); |
7682 | #else | 7674 | #else |
7683 | group = cpu; | 7675 | group = cpu; |
@@ -8020,7 +8012,7 @@ static int __build_sched_domains(const struct cpumask *cpu_map, | |||
8020 | SD_INIT(sd, SIBLING); | 8012 | SD_INIT(sd, SIBLING); |
8021 | set_domain_attribute(sd, attr); | 8013 | set_domain_attribute(sd, attr); |
8022 | cpumask_and(sched_domain_span(sd), | 8014 | cpumask_and(sched_domain_span(sd), |
8023 | &per_cpu(cpu_sibling_map, i), cpu_map); | 8015 | topology_thread_cpumask(i), cpu_map); |
8024 | sd->parent = p; | 8016 | sd->parent = p; |
8025 | p->child = sd; | 8017 | p->child = sd; |
8026 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); | 8018 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
@@ -8031,7 +8023,7 @@ static int __build_sched_domains(const struct cpumask *cpu_map, | |||
8031 | /* Set up CPU (sibling) groups */ | 8023 | /* Set up CPU (sibling) groups */ |
8032 | for_each_cpu(i, cpu_map) { | 8024 | for_each_cpu(i, cpu_map) { |
8033 | cpumask_and(this_sibling_map, | 8025 | cpumask_and(this_sibling_map, |
8034 | &per_cpu(cpu_sibling_map, i), cpu_map); | 8026 | topology_thread_cpumask(i), cpu_map); |
8035 | if (i != cpumask_first(this_sibling_map)) | 8027 | if (i != cpumask_first(this_sibling_map)) |
8036 | continue; | 8028 | continue; |
8037 | 8029 | ||
@@ -8707,6 +8699,9 @@ void __init sched_init(void) | |||
8707 | #ifdef CONFIG_USER_SCHED | 8699 | #ifdef CONFIG_USER_SCHED |
8708 | alloc_size *= 2; | 8700 | alloc_size *= 2; |
8709 | #endif | 8701 | #endif |
8702 | #ifdef CONFIG_CPUMASK_OFFSTACK | ||
8703 | alloc_size += num_possible_cpus() * cpumask_size(); | ||
8704 | #endif | ||
8710 | /* | 8705 | /* |
8711 | * As sched_init() is called before page_alloc is setup, | 8706 | * As sched_init() is called before page_alloc is setup, |
8712 | * we use alloc_bootmem(). | 8707 | * we use alloc_bootmem(). |
@@ -8744,6 +8739,12 @@ void __init sched_init(void) | |||
8744 | ptr += nr_cpu_ids * sizeof(void **); | 8739 | ptr += nr_cpu_ids * sizeof(void **); |
8745 | #endif /* CONFIG_USER_SCHED */ | 8740 | #endif /* CONFIG_USER_SCHED */ |
8746 | #endif /* CONFIG_RT_GROUP_SCHED */ | 8741 | #endif /* CONFIG_RT_GROUP_SCHED */ |
8742 | #ifdef CONFIG_CPUMASK_OFFSTACK | ||
8743 | for_each_possible_cpu(i) { | ||
8744 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | ||
8745 | ptr += cpumask_size(); | ||
8746 | } | ||
8747 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | ||
8747 | } | 8748 | } |
8748 | 8749 | ||
8749 | #ifdef CONFIG_SMP | 8750 | #ifdef CONFIG_SMP |