Update to 3.4-final.
[linux-flexiantxendom0-3.2.10.git] / kernel / sched / core.c
index 985f6e5..9a7fe31 100644 (file)
@@ -71,7 +71,9 @@
 #include <linux/ftrace.h>
 #include <linux/slab.h>
 #include <linux/init_task.h>
+#include <linux/binfmts.h>
 
+#include <asm/switch_to.h>
 #include <asm/tlb.h>
 #include <asm/irq_regs.h>
 #include <asm/mutex.h>
@@ -162,13 +164,13 @@ static int sched_feat_show(struct seq_file *m, void *v)
 
 #ifdef HAVE_JUMP_LABEL
 
-#define jump_label_key__true  jump_label_key_enabled
-#define jump_label_key__false jump_label_key_disabled
+#define jump_label_key__true  STATIC_KEY_INIT_TRUE
+#define jump_label_key__false STATIC_KEY_INIT_FALSE
 
 #define SCHED_FEAT(name, enabled)      \
        jump_label_key__##enabled ,
 
-struct jump_label_key sched_feat_keys[__SCHED_FEAT_NR] = {
+struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
 #include "features.h"
 };
 
@@ -176,14 +178,14 @@ struct jump_label_key sched_feat_keys[__SCHED_FEAT_NR] = {
 
 static void sched_feat_disable(int i)
 {
-       if (jump_label_enabled(&sched_feat_keys[i]))
-               jump_label_dec(&sched_feat_keys[i]);
+       if (static_key_enabled(&sched_feat_keys[i]))
+               static_key_slow_dec(&sched_feat_keys[i]);
 }
 
 static void sched_feat_enable(int i)
 {
-       if (!jump_label_enabled(&sched_feat_keys[i]))
-               jump_label_inc(&sched_feat_keys[i]);
+       if (!static_key_enabled(&sched_feat_keys[i]))
+               static_key_slow_inc(&sched_feat_keys[i]);
 }
 #else
 static void sched_feat_disable(int i) { };
@@ -894,7 +896,7 @@ static void update_rq_clock_task(struct rq *rq, s64 delta)
        delta -= irq_delta;
 #endif
 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
-       if (static_branch((&paravirt_steal_rq_enabled))) {
+       if (static_key_false((&paravirt_steal_rq_enabled))) {
                u64 st;
 
                steal = paravirt_steal_clock(cpu_of(rq));
@@ -1268,7 +1270,7 @@ static int select_fallback_rq(int cpu, struct task_struct *p)
        int dest_cpu;
 
        /* Look for allowed, online CPU in same node. */
-       for_each_cpu_mask(dest_cpu, *nodemask) {
+       for_each_cpu(dest_cpu, nodemask) {
                if (!cpu_online(dest_cpu))
                        continue;
                if (!cpu_active(dest_cpu))
@@ -1279,7 +1281,7 @@ static int select_fallback_rq(int cpu, struct task_struct *p)
 
        for (;;) {
                /* Any allowed, online CPU? */
-               for_each_cpu_mask(dest_cpu, *tsk_cpus_allowed(p)) {
+               for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
                        if (!cpu_online(dest_cpu))
                                continue;
                        if (!cpu_active(dest_cpu))
@@ -2654,6 +2656,48 @@ static inline void task_group_account_field(struct task_struct *p, int index,
 }
 
 
+#if !defined(CONFIG_XEN) || defined(CONFIG_VIRT_CPU_ACCOUNTING)
+# define cputime_to_u64(t) ((__force u64)(t))
+#else
+# include <linux/syscore_ops.h>
+# define NS_PER_TICK (1000000000 / HZ)
+
+static DEFINE_PER_CPU(u64, steal_snapshot);
+static DEFINE_PER_CPU(unsigned int, steal_residual);
+
+static u64 cputime_to_u64(cputime_t t)
+{
+       u64 s = this_vcpu_read(runstate.time[RUNSTATE_runnable]);
+       unsigned long adj = div_u64_rem(s - __this_cpu_read(steal_snapshot)
+                                         + __this_cpu_read(steal_residual),
+                                       NS_PER_TICK,
+                                       &__get_cpu_var(steal_residual));
+
+       __this_cpu_write(steal_snapshot, s);
+       if (t < jiffies_to_cputime(adj))
+               return 0;
+
+       return (__force u64)(t - jiffies_to_cputime(adj));
+}
+
+static void steal_resume(void)
+{
+       cputime_to_u64(((cputime_t)1 << (BITS_PER_LONG * sizeof(cputime_t)
+                                        / sizeof(long) - 1)) - 1);
+}
+
+static struct syscore_ops steal_syscore_ops = {
+       .resume = steal_resume,
+};
+
+static int __init steal_register(void)
+{
+       register_syscore_ops(&steal_syscore_ops);
+       return 0;
+}
+core_initcall(steal_register);
+#endif
+
 /*
  * Account user cpu time to a process.
  * @p: the process that the cpu time gets accounted to
@@ -2673,7 +2717,7 @@ void account_user_time(struct task_struct *p, cputime_t cputime,
        index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
 
        /* Add user time to cpustat. */
-       task_group_account_field(p, index, (__force u64) cputime);
+       task_group_account_field(p, index, cputime_to_u64(cputime));
 
        /* Account for user time used */
        acct_update_integrals(p);
@@ -2723,7 +2767,7 @@ void __account_system_time(struct task_struct *p, cputime_t cputime,
        account_group_system_time(p, cputime);
 
        /* Add system time to cpustat. */
-       task_group_account_field(p, index, (__force u64) cputime);
+       task_group_account_field(p, index, cputime_to_u64(cputime));
 
        /* Account for system time used */
        acct_update_integrals(p);
@@ -2777,15 +2821,15 @@ void account_idle_time(cputime_t cputime)
        struct rq *rq = this_rq();
 
        if (atomic_read(&rq->nr_iowait) > 0)
-               cpustat[CPUTIME_IOWAIT] += (__force u64) cputime;
+               cpustat[CPUTIME_IOWAIT] += cputime_to_u64(cputime);
        else
-               cpustat[CPUTIME_IDLE] += (__force u64) cputime;
+               cpustat[CPUTIME_IDLE] += cputime_to_u64(cputime);
 }
 
 static __always_inline bool steal_account_process_tick(void)
 {
 #ifdef CONFIG_PARAVIRT
-       if (static_branch(&paravirt_steal_enabled)) {
+       if (static_key_false(&paravirt_steal_enabled)) {
                u64 steal, st = 0;
 
                steal = paravirt_steal_clock(smp_processor_id());
@@ -2835,9 +2879,9 @@ static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
                return;
 
        if (irqtime_account_hi_update()) {
-               cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
+               cpustat[CPUTIME_IRQ] += cputime_to_u64(cputime_one_jiffy);
        } else if (irqtime_account_si_update()) {
-               cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
+               cpustat[CPUTIME_SOFTIRQ] += cputime_to_u64(cputime_one_jiffy);
        } else if (this_cpu_ksoftirqd() == p) {
                /*
                 * ksoftirqd time do not get accounted in cpu_softirq_time.
@@ -3283,6 +3327,11 @@ void __sched schedule_preempt_disabled(void)
 }
 
 #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
+#include <asm/mutex.h>
+
+#ifndef arch_cpu_is_running
+#define arch_cpu_is_running(cpu) true
+#endif
 
 static inline bool owner_running(struct mutex *lock, struct task_struct *owner)
 {
@@ -3297,7 +3346,8 @@ static inline bool owner_running(struct mutex *lock, struct task_struct *owner)
         */
        barrier();
 
-       return owner->on_cpu;
+       return owner->on_cpu
+              && arch_cpu_is_running(task_thread_info(owner)->cpu);
 }
 
 /*
@@ -6380,6 +6430,8 @@ static int __sdt_alloc(const struct cpumask *cpu_map)
                        if (!sg)
                                return -ENOMEM;
 
+                       sg->next = sg;
+
                        *per_cpu_ptr(sdd->sg, j) = sg;
 
                        sgp = kzalloc_node(sizeof(struct sched_group_power),
@@ -6403,16 +6455,26 @@ static void __sdt_free(const struct cpumask *cpu_map)
                struct sd_data *sdd = &tl->data;
 
                for_each_cpu(j, cpu_map) {
-                       struct sched_domain *sd = *per_cpu_ptr(sdd->sd, j);
-                       if (sd && (sd->flags & SD_OVERLAP))
-                               free_sched_groups(sd->groups, 0);
-                       kfree(*per_cpu_ptr(sdd->sd, j));
-                       kfree(*per_cpu_ptr(sdd->sg, j));
-                       kfree(*per_cpu_ptr(sdd->sgp, j));
+                       struct sched_domain *sd;
+
+                       if (sdd->sd) {
+                               sd = *per_cpu_ptr(sdd->sd, j);
+                               if (sd && (sd->flags & SD_OVERLAP))
+                                       free_sched_groups(sd->groups, 0);
+                               kfree(*per_cpu_ptr(sdd->sd, j));
+                       }
+
+                       if (sdd->sg)
+                               kfree(*per_cpu_ptr(sdd->sg, j));
+                       if (sdd->sgp)
+                               kfree(*per_cpu_ptr(sdd->sgp, j));
                }
                free_percpu(sdd->sd);
+               sdd->sd = NULL;
                free_percpu(sdd->sg);
+               sdd->sg = NULL;
                free_percpu(sdd->sgp);
+               sdd->sgp = NULL;
        }
 }
 
@@ -7596,8 +7658,7 @@ static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
                            struct task_group, css);
 }
 
-static struct cgroup_subsys_state *
-cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
+static struct cgroup_subsys_state *cpu_cgroup_create(struct cgroup *cgrp)
 {
        struct task_group *tg, *parent;
 
@@ -7614,15 +7675,14 @@ cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp)
        return &tg->css;
 }
 
-static void
-cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
+static void cpu_cgroup_destroy(struct cgroup *cgrp)
 {
        struct task_group *tg = cgroup_tg(cgrp);
 
        sched_destroy_group(tg);
 }
 
-static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
+static int cpu_cgroup_can_attach(struct cgroup *cgrp,
                                 struct cgroup_taskset *tset)
 {
        struct task_struct *task;
@@ -7640,7 +7700,7 @@ static int cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
        return 0;
 }
 
-static void cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
+static void cpu_cgroup_attach(struct cgroup *cgrp,
                              struct cgroup_taskset *tset)
 {
        struct task_struct *task;
@@ -7650,8 +7710,8 @@ static void cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
 }
 
 static void
-cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp,
-               struct cgroup *old_cgrp, struct task_struct *task)
+cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp,
+               struct task_struct *task)
 {
        /*
         * cgroup_exit() is called in the copy_process() failure path.
@@ -8001,8 +8061,7 @@ struct cgroup_subsys cpu_cgroup_subsys = {
  */
 
 /* create a new cpu accounting group */
-static struct cgroup_subsys_state *cpuacct_create(
-       struct cgroup_subsys *ss, struct cgroup *cgrp)
+static struct cgroup_subsys_state *cpuacct_create(struct cgroup *cgrp)
 {
        struct cpuacct *ca;
 
@@ -8032,8 +8091,7 @@ out:
 }
 
 /* destroy an existing cpu accounting group */
-static void
-cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
+static void cpuacct_destroy(struct cgroup *cgrp)
 {
        struct cpuacct *ca = cgroup_ca(cgrp);