#include <linux/latencytop.h>
#include <linux/sched.h>
#include <linux/cpumask.h>
+#include <linux/cpuidle.h>
#include <linux/slab.h>
#include <linux/profile.h>
#include <linux/interrupt.h>
static unsigned int task_scan_min(struct task_struct *p)
{
+ unsigned int scan_size = ACCESS_ONCE(sysctl_numa_balancing_scan_size);
unsigned int scan, floor;
unsigned int windows = 1;
- if (sysctl_numa_balancing_scan_size < MAX_SCAN_WINDOW)
- windows = MAX_SCAN_WINDOW / sysctl_numa_balancing_scan_size;
+ if (scan_size < MAX_SCAN_WINDOW)
+ windows = MAX_SCAN_WINDOW / scan_size;
floor = 1000 / windows;
scan = sysctl_numa_balancing_scan_period_min / task_nr_scan_windows(p);
group->faults_cpu[task_faults_idx(nid, 1)];
}
+/* Handle placement on systems where not all nodes are directly connected. */
+static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
+ int maxdist, bool task)
+{
+ unsigned long score = 0;
+ int node;
+
+ /*
+ * All nodes are directly connected, and the same distance
+ * from each other. No need for fancy placement algorithms.
+ */
+ if (sched_numa_topology_type == NUMA_DIRECT)
+ return 0;
+
+ /*
+ * This code is called for each node, introducing N^2 complexity,
+ * which should be ok given the number of nodes rarely exceeds 8.
+ */
+ for_each_online_node(node) {
+ unsigned long faults;
+ int dist = node_distance(nid, node);
+
+ /*
+ * The furthest away nodes in the system are not interesting
+ * for placement; nid was already counted.
+ */
+ if (dist == sched_max_numa_distance || node == nid)
+ continue;
+
+ /*
+ * On systems with a backplane NUMA topology, compare groups
+ * of nodes, and move tasks towards the group with the most
+ * memory accesses. When comparing two nodes at distance
+ * "hoplimit", only nodes closer by than "hoplimit" are part
+ * of each group. Skip other nodes.
+ */
+ if (sched_numa_topology_type == NUMA_BACKPLANE &&
+ dist > maxdist)
+ continue;
+
+ /* Add up the faults from nearby nodes. */
+ if (task)
+ faults = task_faults(p, node);
+ else
+ faults = group_faults(p, node);
+
+ /*
+ * On systems with a glueless mesh NUMA topology, there are
+ * no fixed "groups of nodes". Instead, nodes that are not
+ * directly connected bounce traffic through intermediate
+ * nodes; a numa_group can occupy any set of nodes.
+ * The further away a node is, the less the faults count.
+ * This seems to result in good task placement.
+ */
+ if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
+ faults *= (sched_max_numa_distance - dist);
+ faults /= (sched_max_numa_distance - LOCAL_DISTANCE);
+ }
+
+ score += faults;
+ }
+
+ return score;
+}
+
/*
* These return the fraction of accesses done by a particular task, or
* task group, on a particular numa node. The group weight is given a
* larger multiplier, in order to group tasks together that are almost
* evenly spread out between numa nodes.
*/
-static inline unsigned long task_weight(struct task_struct *p, int nid)
+static inline unsigned long task_weight(struct task_struct *p, int nid,
+ int dist)
{
- unsigned long total_faults;
+ unsigned long faults, total_faults;
if (!p->numa_faults_memory)
return 0;
if (!total_faults)
return 0;
- return 1000 * task_faults(p, nid) / total_faults;
+ faults = task_faults(p, nid);
+ faults += score_nearby_nodes(p, nid, dist, true);
+
+ return 1000 * faults / total_faults;
}
-static inline unsigned long group_weight(struct task_struct *p, int nid)
+static inline unsigned long group_weight(struct task_struct *p, int nid,
+ int dist)
{
- if (!p->numa_group || !p->numa_group->total_faults)
+ unsigned long faults, total_faults;
+
+ if (!p->numa_group)
return 0;
- return 1000 * group_faults(p, nid) / p->numa_group->total_faults;
+ total_faults = p->numa_group->total_faults;
+
+ if (!total_faults)
+ return 0;
+
+ faults = group_faults(p, nid);
+ faults += score_nearby_nodes(p, nid, dist, false);
+
+ return 1000 * faults / total_faults;
}
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
struct numa_stats src_stats, dst_stats;
int imbalance_pct;
+ int dist;
struct task_struct *best_task;
long best_imp;
long load;
long imp = env->p->numa_group ? groupimp : taskimp;
long moveimp = imp;
+ int dist = env->dist;
rcu_read_lock();
- cur = ACCESS_ONCE(dst_rq->curr);
- if (cur->pid == 0) /* idle */
+
+ raw_spin_lock_irq(&dst_rq->lock);
+ cur = dst_rq->curr;
+ /*
+ * No need to move the exiting task, and this ensures that ->curr
+ * wasn't reaped and thus get_task_struct() in task_numa_assign()
+ * is safe under RCU read lock.
+ * Note that rcu_read_lock() itself can't protect from the final
+ * put_task_struct() after the last schedule().
+ */
+ if ((cur->flags & PF_EXITING) || is_idle_task(cur))
cur = NULL;
+ raw_spin_unlock_irq(&dst_rq->lock);
/*
* "imp" is the fault differential for the source task between the
* in any group then look only at task weights.
*/
if (cur->numa_group == env->p->numa_group) {
- imp = taskimp + task_weight(cur, env->src_nid) -
- task_weight(cur, env->dst_nid);
+ imp = taskimp + task_weight(cur, env->src_nid, dist) -
+ task_weight(cur, env->dst_nid, dist);
/*
* Add some hysteresis to prevent swapping the
* tasks within a group over tiny differences.
* instead.
*/
if (cur->numa_group)
- imp += group_weight(cur, env->src_nid) -
- group_weight(cur, env->dst_nid);
+ imp += group_weight(cur, env->src_nid, dist) -
+ group_weight(cur, env->dst_nid, dist);
else
- imp += task_weight(cur, env->src_nid) -
- task_weight(cur, env->dst_nid);
+ imp += task_weight(cur, env->src_nid, dist) -
+ task_weight(cur, env->dst_nid, dist);
}
}
};
struct sched_domain *sd;
unsigned long taskweight, groupweight;
- int nid, ret;
+ int nid, ret, dist;
long taskimp, groupimp;
/*
return -EINVAL;
}
- taskweight = task_weight(p, env.src_nid);
- groupweight = group_weight(p, env.src_nid);
- update_numa_stats(&env.src_stats, env.src_nid);
env.dst_nid = p->numa_preferred_nid;
- taskimp = task_weight(p, env.dst_nid) - taskweight;
- groupimp = group_weight(p, env.dst_nid) - groupweight;
+ dist = env.dist = node_distance(env.src_nid, env.dst_nid);
+ taskweight = task_weight(p, env.src_nid, dist);
+ groupweight = group_weight(p, env.src_nid, dist);
+ update_numa_stats(&env.src_stats, env.src_nid);
+ taskimp = task_weight(p, env.dst_nid, dist) - taskweight;
+ groupimp = group_weight(p, env.dst_nid, dist) - groupweight;
update_numa_stats(&env.dst_stats, env.dst_nid);
/* Try to find a spot on the preferred nid. */
if (nid == env.src_nid || nid == p->numa_preferred_nid)
continue;
+ dist = node_distance(env.src_nid, env.dst_nid);
+ if (sched_numa_topology_type == NUMA_BACKPLANE &&
+ dist != env.dist) {
+ taskweight = task_weight(p, env.src_nid, dist);
+ groupweight = group_weight(p, env.src_nid, dist);
+ }
+
/* Only consider nodes where both task and groups benefit */
- taskimp = task_weight(p, nid) - taskweight;
- groupimp = group_weight(p, nid) - groupweight;
+ taskimp = task_weight(p, nid, dist) - taskweight;
+ groupimp = group_weight(p, nid, dist) - groupweight;
if (taskimp < 0 && groupimp < 0)
continue;
+ env.dist = dist;
env.dst_nid = nid;
update_numa_stats(&env.dst_stats, env.dst_nid);
task_numa_find_cpu(&env, taskimp, groupimp);
* scanning faster if shared accesses dominate as it may
* simply bounce migrations uselessly
*/
- ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared));
+ ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
diff = (diff * ratio) / NUMA_PERIOD_SLOTS;
}
return delta;
}
+/*
+ * Determine the preferred nid for a task in a numa_group. This needs to
+ * be done in a way that produces consistent results with group_weight,
+ * otherwise workloads might not converge.
+ */
+static int preferred_group_nid(struct task_struct *p, int nid)
+{
+ nodemask_t nodes;
+ int dist;
+
+ /* Direct connections between all NUMA nodes. */
+ if (sched_numa_topology_type == NUMA_DIRECT)
+ return nid;
+
+ /*
+ * On a system with glueless mesh NUMA topology, group_weight
+ * scores nodes according to the number of NUMA hinting faults on
+ * both the node itself, and on nearby nodes.
+ */
+ if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
+ unsigned long score, max_score = 0;
+ int node, max_node = nid;
+
+ dist = sched_max_numa_distance;
+
+ for_each_online_node(node) {
+ score = group_weight(p, node, dist);
+ if (score > max_score) {
+ max_score = score;
+ max_node = node;
+ }
+ }
+ return max_node;
+ }
+
+ /*
+ * Finding the preferred nid in a system with NUMA backplane
+ * interconnect topology is more involved. The goal is to locate
+ * tasks from numa_groups near each other in the system, and
+ * untangle workloads from different sides of the system. This requires
+ * searching down the hierarchy of node groups, recursively searching
+ * inside the highest scoring group of nodes. The nodemask tricks
+ * keep the complexity of the search down.
+ */
+ nodes = node_online_map;
+ for (dist = sched_max_numa_distance; dist > LOCAL_DISTANCE; dist--) {
+ unsigned long max_faults = 0;
+ nodemask_t max_group;
+ int a, b;
+
+ /* Are there nodes at this distance from each other? */
+ if (!find_numa_distance(dist))
+ continue;
+
+ for_each_node_mask(a, nodes) {
+ unsigned long faults = 0;
+ nodemask_t this_group;
+ nodes_clear(this_group);
+
+ /* Sum group's NUMA faults; includes a==b case. */
+ for_each_node_mask(b, nodes) {
+ if (node_distance(a, b) < dist) {
+ faults += group_faults(p, b);
+ node_set(b, this_group);
+ node_clear(b, nodes);
+ }
+ }
+
+ /* Remember the top group. */
+ if (faults > max_faults) {
+ max_faults = faults;
+ max_group = this_group;
+ /*
+ * subtle: at the smallest distance there is
+ * just one node left in each "group", the
+ * winner is the preferred nid.
+ */
+ nid = a;
+ }
+ }
+ /* Next round, evaluate the nodes within max_group. */
+ nodes = max_group;
+ }
+ return nid;
+}
+
static void task_numa_placement(struct task_struct *p)
{
int seq, nid, max_nid = -1, max_group_nid = -1;
if (p->numa_group) {
update_numa_active_node_mask(p->numa_group);
spin_unlock_irq(group_lock);
- max_nid = max_group_nid;
+ max_nid = preferred_group_nid(p, max_group_nid);
}
if (max_faults) {
if (!p->mm)
return;
- /* Do not worry about placement if exiting */
- if (p->state == TASK_DEAD)
- return;
-
/* Allocate buffer to track faults on a per-node basis */
if (unlikely(!p->numa_faults_memory)) {
int size = sizeof(*p->numa_faults_memory) *
vma = mm->mmap;
}
for (; vma; vma = vma->vm_next) {
- if (!vma_migratable(vma) || !vma_policy_mof(p, vma))
+ if (!vma_migratable(vma) || !vma_policy_mof(vma))
continue;
/*
/*
* As y^PERIOD = 1/2, we can combine
- * y^n = 1/2^(n/PERIOD) * k^(n%PERIOD)
- * With a look-up table which covers k^n (n<PERIOD)
+ * y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
+ * With a look-up table which covers y^n (n<PERIOD)
*
* To achieve constant time decay_load.
*/
static unsigned long cpu_avg_load_per_task(int cpu)
{
struct rq *rq = cpu_rq(cpu);
- unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
+ unsigned long nr_running = ACCESS_ONCE(rq->cfs.h_nr_running);
unsigned long load_avg = rq->cfs.runnable_load_avg;
if (nr_running)
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
{
s64 this_load, load;
+ s64 this_eff_load, prev_eff_load;
int idx, this_cpu, prev_cpu;
- unsigned long tl_per_task;
struct task_group *tg;
unsigned long weight;
int balanced;
* Otherwise check if either cpus are near enough in load to allow this
* task to be woken on this_cpu.
*/
- if (this_load > 0) {
- s64 this_eff_load, prev_eff_load;
+ this_eff_load = 100;
+ this_eff_load *= capacity_of(prev_cpu);
+
+ prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
+ prev_eff_load *= capacity_of(this_cpu);
- this_eff_load = 100;
- this_eff_load *= capacity_of(prev_cpu);
+ if (this_load > 0) {
this_eff_load *= this_load +
effective_load(tg, this_cpu, weight, weight);
- prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
- prev_eff_load *= capacity_of(this_cpu);
prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
+ }
- balanced = this_eff_load <= prev_eff_load;
- } else
- balanced = true;
-
- /*
- * If the currently running task will sleep within
- * a reasonable amount of time then attract this newly
- * woken task:
- */
- if (sync && balanced)
- return 1;
+ balanced = this_eff_load <= prev_eff_load;
schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
- tl_per_task = cpu_avg_load_per_task(this_cpu);
- if (balanced ||
- (this_load <= load &&
- this_load + target_load(prev_cpu, idx) <= tl_per_task)) {
- /*
- * This domain has SD_WAKE_AFFINE and
- * p is cache cold in this domain, and
- * there is no bad imbalance.
- */
- schedstat_inc(sd, ttwu_move_affine);
- schedstat_inc(p, se.statistics.nr_wakeups_affine);
+ if (!balanced)
+ return 0;
- return 1;
- }
- return 0;
+ schedstat_inc(sd, ttwu_move_affine);
+ schedstat_inc(p, se.statistics.nr_wakeups_affine);
+
+ return 1;
}
/*
find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
{
unsigned long load, min_load = ULONG_MAX;
- int idlest = -1;
+ unsigned int min_exit_latency = UINT_MAX;
+ u64 latest_idle_timestamp = 0;
+ int least_loaded_cpu = this_cpu;
+ int shallowest_idle_cpu = -1;
int i;
/* Traverse only the allowed CPUs */
for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
- load = weighted_cpuload(i);
-
- if (load < min_load || (load == min_load && i == this_cpu)) {
- min_load = load;
- idlest = i;
+ if (idle_cpu(i)) {
+ struct rq *rq = cpu_rq(i);
+ struct cpuidle_state *idle = idle_get_state(rq);
+ if (idle && idle->exit_latency < min_exit_latency) {
+ /*
+ * We give priority to a CPU whose idle state
+ * has the smallest exit latency irrespective
+ * of any idle timestamp.
+ */
+ min_exit_latency = idle->exit_latency;
+ latest_idle_timestamp = rq->idle_stamp;
+ shallowest_idle_cpu = i;
+ } else if ((!idle || idle->exit_latency == min_exit_latency) &&
+ rq->idle_stamp > latest_idle_timestamp) {
+ /*
+ * If equal or no active idle state, then
+ * the most recently idled CPU might have
+ * a warmer cache.
+ */
+ latest_idle_timestamp = rq->idle_stamp;
+ shallowest_idle_cpu = i;
+ }
+ } else {
+ load = weighted_cpuload(i);
+ if (load < min_load || (load == min_load && i == this_cpu)) {
+ min_load = load;
+ least_loaded_cpu = i;
+ }
}
}
- return idlest;
+ return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
}
/*
if (!tsk_cache_hot)
tsk_cache_hot = migrate_degrades_locality(p, env);
- if (migrate_improves_locality(p, env)) {
-#ifdef CONFIG_SCHEDSTATS
- if (tsk_cache_hot) {
- schedstat_inc(env->sd, lb_hot_gained[env->idle]);
- schedstat_inc(p, se.statistics.nr_forced_migrations);
- }
-#endif
- return 1;
- }
-
- if (!tsk_cache_hot ||
- env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
-
+ if (migrate_improves_locality(p, env) || !tsk_cache_hot ||
+ env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
if (tsk_cache_hot) {
schedstat_inc(env->sd, lb_hot_gained[env->idle]);
schedstat_inc(p, se.statistics.nr_forced_migrations);
}
-
return 1;
}
return default_scale_capacity(sd, cpu);
}
-static unsigned long default_scale_smt_capacity(struct sched_domain *sd, int cpu)
+static unsigned long default_scale_cpu_capacity(struct sched_domain *sd, int cpu)
{
- unsigned long weight = sd->span_weight;
- unsigned long smt_gain = sd->smt_gain;
-
- smt_gain /= weight;
+ if ((sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1))
+ return sd->smt_gain / sd->span_weight;
- return smt_gain;
+ return SCHED_CAPACITY_SCALE;
}
-unsigned long __weak arch_scale_smt_capacity(struct sched_domain *sd, int cpu)
+unsigned long __weak arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
{
- return default_scale_smt_capacity(sd, cpu);
+ return default_scale_cpu_capacity(sd, cpu);
}
static unsigned long scale_rt_capacity(int cpu)
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
{
- unsigned long weight = sd->span_weight;
unsigned long capacity = SCHED_CAPACITY_SCALE;
struct sched_group *sdg = sd->groups;
- if ((sd->flags & SD_SHARE_CPUCAPACITY) && weight > 1) {
- if (sched_feat(ARCH_CAPACITY))
- capacity *= arch_scale_smt_capacity(sd, cpu);
- else
- capacity *= default_scale_smt_capacity(sd, cpu);
+ if (sched_feat(ARCH_CAPACITY))
+ capacity *= arch_scale_cpu_capacity(sd, cpu);
+ else
+ capacity *= default_scale_cpu_capacity(sd, cpu);
- capacity >>= SCHED_CAPACITY_SHIFT;
- }
+ capacity >>= SCHED_CAPACITY_SHIFT;
sdg->sgc->capacity_orig = capacity;
load = source_load(i, load_idx);
sgs->group_load += load;
- sgs->sum_nr_running += rq->nr_running;
+ sgs->sum_nr_running += rq->cfs.h_nr_running;
if (rq->nr_running > 1)
*overload = true;
goto force_balance;
/*
- * If the local group is more busy than the selected busiest group
+ * If the local group is busier than the selected busiest group
* don't try and pull any tasks.
*/
if (local->avg_load >= busiest->avg_load)
if (env->idle == CPU_IDLE) {
/*
- * This cpu is idle. If the busiest group load doesn't
- * have more tasks than the number of available cpu's and
- * there is no imbalance between this and busiest group
- * wrt to idle cpu's, it is balanced.
+ * This cpu is idle. If the busiest group is not overloaded
+ * and there is no imbalance between this and busiest group
+ * wrt idle cpus, it is balanced. The imbalance becomes
+ * significant if the diff is greater than 1 otherwise we
+ * might end up to just move the imbalance on another group
*/
- if ((local->idle_cpus < busiest->idle_cpus) &&
- busiest->sum_nr_running <= busiest->group_weight)
+ if ((busiest->group_type != group_overloaded) &&
+ (local->idle_cpus <= (busiest->idle_cpus + 1)))
goto out_balanced;
} else {
/*
struct sched_group *group;
struct rq *busiest;
unsigned long flags;
- struct cpumask *cpus = __get_cpu_var(load_balance_mask);
+ struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
struct lb_env env = {
.sd = sd,
local_irq_restore(flags);
- /*
- * some other cpu did the load balance for us.
- */
- if (cur_ld_moved && env.dst_cpu != smp_processor_id())
- resched_cpu(env.dst_cpu);
-
if (env.flags & LBF_NEED_BREAK) {
env.flags &= ~LBF_NEED_BREAK;
goto more_balance;
if (sd_parent) {
int *group_imbalance = &sd_parent->groups->sgc->imbalance;
- if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0) {
+ if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
*group_imbalance = 1;
- } else if (*group_imbalance)
- *group_imbalance = 0;
}
/* All tasks on this runqueue were pinned by CPU affinity */
env.loop_break = sched_nr_migrate_break;
goto redo;
}
- goto out_balanced;
+ goto out_all_pinned;
}
}
goto out;
out_balanced:
+ /*
+ * We reach balance although we may have faced some affinity
+ * constraints. Clear the imbalance flag if it was set.
+ */
+ if (sd_parent) {
+ int *group_imbalance = &sd_parent->groups->sgc->imbalance;
+
+ if (*group_imbalance)
+ *group_imbalance = 0;
+ }
+
+out_all_pinned:
+ /*
+ * We reach balance because all tasks are pinned at this level so
+ * we can't migrate them. Let the imbalance flag set so parent level
+ * can try to migrate them.
+ */
schedstat_inc(sd, lb_balanced[idle]);
sd->nr_balance_failed = 0;