CGroups _s64 files: use read_s64/write_s64 in CFS cgroup for rt_runtime file
[safe/jmp/linux-2.6] / kernel / sched.c
index 6d55dfc..e2f7f5a 100644 (file)
@@ -68,6 +68,8 @@
 #include <linux/hrtimer.h>
 #include <linux/tick.h>
 #include <linux/bootmem.h>
+#include <linux/debugfs.h>
+#include <linux/ctype.h>
 
 #include <asm/tlb.h>
 #include <asm/irq_regs.h>
@@ -732,30 +734,137 @@ static void update_rq_clock(struct rq *rq)
 /*
  * Debugging: various feature bits
  */
+
+#define SCHED_FEAT(name, enabled)      \
+       __SCHED_FEAT_##name ,
+
 enum {
-       SCHED_FEAT_NEW_FAIR_SLEEPERS    = 1,
-       SCHED_FEAT_WAKEUP_PREEMPT       = 2,
-       SCHED_FEAT_START_DEBIT          = 4,
-       SCHED_FEAT_AFFINE_WAKEUPS       = 8,
-       SCHED_FEAT_CACHE_HOT_BUDDY      = 16,
-       SCHED_FEAT_SYNC_WAKEUPS         = 32,
-       SCHED_FEAT_HRTICK               = 64,
-       SCHED_FEAT_DOUBLE_TICK          = 128,
-       SCHED_FEAT_NORMALIZED_SLEEPER   = 256,
+#include "sched_features.h"
 };
 
+#undef SCHED_FEAT
+
+#define SCHED_FEAT(name, enabled)      \
+       (1UL << __SCHED_FEAT_##name) * enabled |
+
 const_debug unsigned int sysctl_sched_features =
-               SCHED_FEAT_NEW_FAIR_SLEEPERS    * 1 |
-               SCHED_FEAT_WAKEUP_PREEMPT       * 1 |
-               SCHED_FEAT_START_DEBIT          * 1 |
-               SCHED_FEAT_AFFINE_WAKEUPS       * 1 |
-               SCHED_FEAT_CACHE_HOT_BUDDY      * 1 |
-               SCHED_FEAT_SYNC_WAKEUPS         * 1 |
-               SCHED_FEAT_HRTICK               * 1 |
-               SCHED_FEAT_DOUBLE_TICK          * 0 |
-               SCHED_FEAT_NORMALIZED_SLEEPER   * 1;
+#include "sched_features.h"
+       0;
+
+#undef SCHED_FEAT
+
+#ifdef CONFIG_SCHED_DEBUG
+#define SCHED_FEAT(name, enabled)      \
+       #name ,
+
+__read_mostly char *sched_feat_names[] = {
+#include "sched_features.h"
+       NULL
+};
+
+#undef SCHED_FEAT
+
+int sched_feat_open(struct inode *inode, struct file *filp)
+{
+       filp->private_data = inode->i_private;
+       return 0;
+}
+
+static ssize_t
+sched_feat_read(struct file *filp, char __user *ubuf,
+               size_t cnt, loff_t *ppos)
+{
+       char *buf;
+       int r = 0;
+       int len = 0;
+       int i;
+
+       for (i = 0; sched_feat_names[i]; i++) {
+               len += strlen(sched_feat_names[i]);
+               len += 4;
+       }
+
+       buf = kmalloc(len + 2, GFP_KERNEL);
+       if (!buf)
+               return -ENOMEM;
+
+       for (i = 0; sched_feat_names[i]; i++) {
+               if (sysctl_sched_features & (1UL << i))
+                       r += sprintf(buf + r, "%s ", sched_feat_names[i]);
+               else
+                       r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]);
+       }
 
-#define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x)
+       r += sprintf(buf + r, "\n");
+       WARN_ON(r >= len + 2);
+
+       r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
+
+       kfree(buf);
+
+       return r;
+}
+
+static ssize_t
+sched_feat_write(struct file *filp, const char __user *ubuf,
+               size_t cnt, loff_t *ppos)
+{
+       char buf[64];
+       char *cmp = buf;
+       int neg = 0;
+       int i;
+
+       if (cnt > 63)
+               cnt = 63;
+
+       if (copy_from_user(&buf, ubuf, cnt))
+               return -EFAULT;
+
+       buf[cnt] = 0;
+
+       if (strncmp(buf, "NO_", 3) == 0) {
+               neg = 1;
+               cmp += 3;
+       }
+
+       for (i = 0; sched_feat_names[i]; i++) {
+               int len = strlen(sched_feat_names[i]);
+
+               if (strncmp(cmp, sched_feat_names[i], len) == 0) {
+                       if (neg)
+                               sysctl_sched_features &= ~(1UL << i);
+                       else
+                               sysctl_sched_features |= (1UL << i);
+                       break;
+               }
+       }
+
+       if (!sched_feat_names[i])
+               return -EINVAL;
+
+       filp->f_pos += cnt;
+
+       return cnt;
+}
+
+static struct file_operations sched_feat_fops = {
+       .open   = sched_feat_open,
+       .read   = sched_feat_read,
+       .write  = sched_feat_write,
+};
+
+static __init int sched_init_debug(void)
+{
+       debugfs_create_file("sched_features", 0644, NULL, NULL,
+                       &sched_feat_fops);
+
+       return 0;
+}
+late_initcall(sched_init_debug);
+
+#endif
+
+#define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x))
 
 /*
  * Number of tasks to iterate in a single balance run.
@@ -1548,42 +1657,6 @@ void aggregate_group_weight(struct task_group *tg, struct sched_domain *sd)
 }
 
 /*
- * Redistribute tg->shares amongst all tg->cfs_rq[]s.
- */
-static void __aggregate_redistribute_shares(struct task_group *tg)
-{
-       int i, max_cpu = smp_processor_id();
-       unsigned long rq_weight = 0;
-       unsigned long shares, max_shares = 0, shares_rem = tg->shares;
-
-       for_each_possible_cpu(i)
-               rq_weight += tg->cfs_rq[i]->load.weight;
-
-       for_each_possible_cpu(i) {
-               /*
-                * divide shares proportional to the rq_weights.
-                */
-               shares = tg->shares * tg->cfs_rq[i]->load.weight;
-               shares /= rq_weight + 1;
-
-               tg->cfs_rq[i]->shares = shares;
-
-               if (shares > max_shares) {
-                       max_shares = shares;
-                       max_cpu = i;
-               }
-               shares_rem -= shares;
-       }
-
-       /*
-        * Ensure it all adds up to tg->shares; we can loose a few
-        * due to rounding down when computing the per-cpu shares.
-        */
-       if (shares_rem)
-               tg->cfs_rq[max_cpu]->shares += shares_rem;
-}
-
-/*
  * Compute the weight of this group on the given cpus.
  */
 static
@@ -1592,18 +1665,11 @@ void aggregate_group_shares(struct task_group *tg, struct sched_domain *sd)
        unsigned long shares = 0;
        int i;
 
-again:
        for_each_cpu_mask(i, sd->span)
                shares += tg->cfs_rq[i]->shares;
 
-       /*
-        * When the span doesn't have any shares assigned, but does have
-        * tasks to run do a machine wide rebalance (should be rare).
-        */
-       if (unlikely(!shares && aggregate(tg, sd)->rq_weight)) {
-               __aggregate_redistribute_shares(tg);
-               goto again;
-       }
+       if ((!shares && aggregate(tg, sd)->rq_weight) || shares > tg->shares)
+               shares = tg->shares;
 
        aggregate(tg, sd)->shares = shares;
 }
@@ -6926,6 +6992,7 @@ static int find_next_best_node(int node, nodemask_t *used_nodes)
 /**
  * sched_domain_node_span - get a cpumask for a node's sched_domain
  * @node: node whose cpumask we're constructing
+ * @span: resulting cpumask
  *
  * Given a node, construct a good cpumask for its sched_domain to span. It
  * should be one that prevents unnecessary balancing, but also spreads tasks
@@ -7881,11 +7948,6 @@ void __init sched_init_smp(void)
 #else
 void __init sched_init_smp(void)
 {
-#if defined(CONFIG_NUMA)
-       sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
-                                                               GFP_KERNEL);
-       BUG_ON(sched_group_nodes_bycpu == NULL);
-#endif
        sched_init_granularity();
 }
 #endif /* CONFIG_SMP */
@@ -8018,7 +8080,7 @@ void __init sched_init(void)
         * we use alloc_bootmem().
         */
        if (alloc_size) {
-               ptr = (unsigned long)alloc_bootmem_low(alloc_size);
+               ptr = (unsigned long)alloc_bootmem(alloc_size);
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
                init_task_group.se = (struct sched_entity **)ptr;
@@ -8995,13 +9057,13 @@ cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
 }
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
-static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype,
+static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
                                u64 shareval)
 {
        return sched_group_set_shares(cgroup_tg(cgrp), shareval);
 }
 
-static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft)
+static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
 {
        struct task_group *tg = cgroup_tg(cgrp);
 
@@ -9011,48 +9073,14 @@ static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft)
 
 #ifdef CONFIG_RT_GROUP_SCHED
 static ssize_t cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
-                               struct file *file,
-                               const char __user *userbuf,
-                               size_t nbytes, loff_t *unused_ppos)
+                               s64 val)
 {
-       char buffer[64];
-       int retval = 0;
-       s64 val;
-       char *end;
-
-       if (!nbytes)
-               return -EINVAL;
-       if (nbytes >= sizeof(buffer))
-               return -E2BIG;
-       if (copy_from_user(buffer, userbuf, nbytes))
-               return -EFAULT;
-
-       buffer[nbytes] = 0;     /* nul-terminate */
-
-       /* strip newline if necessary */
-       if (nbytes && (buffer[nbytes-1] == '\n'))
-               buffer[nbytes-1] = 0;
-       val = simple_strtoll(buffer, &end, 0);
-       if (*end)
-               return -EINVAL;
-
-       /* Pass to subsystem */
-       retval = sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
-       if (!retval)
-               retval = nbytes;
-       return retval;
+       return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
 }
 
-static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft,
-                                  struct file *file,
-                                  char __user *buf, size_t nbytes,
-                                  loff_t *ppos)
+static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
 {
-       char tmp[64];
-       long val = sched_group_rt_runtime(cgroup_tg(cgrp));
-       int len = sprintf(tmp, "%ld\n", val);
-
-       return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
+       return sched_group_rt_runtime(cgroup_tg(cgrp));
 }
 
 static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
@@ -9071,20 +9099,20 @@ static struct cftype cpu_files[] = {
 #ifdef CONFIG_FAIR_GROUP_SCHED
        {
                .name = "shares",
-               .read_uint = cpu_shares_read_uint,
-               .write_uint = cpu_shares_write_uint,
+               .read_u64 = cpu_shares_read_u64,
+               .write_u64 = cpu_shares_write_u64,
        },
 #endif
 #ifdef CONFIG_RT_GROUP_SCHED
        {
                .name = "rt_runtime_us",
-               .read = cpu_rt_runtime_read,
-               .write = cpu_rt_runtime_write,
+               .read_s64 = cpu_rt_runtime_read,
+               .write_s64 = cpu_rt_runtime_write,
        },
        {
                .name = "rt_period_us",
-               .read_uint = cpu_rt_period_read_uint,
-               .write_uint = cpu_rt_period_write_uint,
+               .read_u64 = cpu_rt_period_read_uint,
+               .write_u64 = cpu_rt_period_write_uint,
        },
 #endif
 };
@@ -9215,8 +9243,8 @@ out:
 static struct cftype files[] = {
        {
                .name = "usage",
-               .read_uint = cpuusage_read,
-               .write_uint = cpuusage_write,
+               .read_u64 = cpuusage_read,
+               .write_u64 = cpuusage_write,
        },
 };