reflect cpupool in numa node affinity

In order to prefer node local memory for a domain the numa node
locality info must be built according to the cpus belonging to the
cpupool of the domain.

Signed-off-by: juergen.gross@ts.fujitsu.com
Committed-by: Keir Fraser <keir@xen.org>

xen/common/cpupool.c

@@ -220,6 +220,7 @@ static int cpupool_assign_cpu_locked(struct cpupool *c, unsigned int cpu)
 {
     int ret;
     struct cpupool *old;
+    struct domain *d;
 
     if ( (cpupool_moving_cpu == cpu) && (c != cpupool_cpu_moving) )
         return -EBUSY;
@@ -240,6 +241,14 @@ static int cpupool_assign_cpu_locked(struct cpupool *c, unsigned int cpu)
         cpupool_cpu_moving = NULL;
     }
     cpumask_set_cpu(cpu, c->cpu_valid);
+
+    rcu_read_lock(&domlist_read_lock);
+    for_each_domain_in_cpupool(d, c)
+    {
+        domain_update_node_affinity(d);
+    }
+    rcu_read_unlock(&domlist_read_lock);
+
     return 0;
 }
 

xen/common/domain.c

@@ -11,6 +11,7 @@
 #include <xen/ctype.h>
 #include <xen/errno.h>
 #include <xen/sched.h>
+#include <xen/sched-if.h>
 #include <xen/domain.h>
 #include <xen/mm.h>
 #include <xen/event.h>
@@ -335,17 +336,29 @@ struct domain *domain_create(
 void domain_update_node_affinity(struct domain *d)
 {
     cpumask_var_t cpumask;
+    cpumask_var_t online_affinity;
+    const cpumask_t *online;
     nodemask_t nodemask = NODE_MASK_NONE;
     struct vcpu *v;
     unsigned int node;
 
     if ( !zalloc_cpumask_var(&cpumask) )
         return;
+    if ( !alloc_cpumask_var(&online_affinity) )
+    {
+        free_cpumask_var(cpumask);
+        return;
+    }
+
+    online = cpupool_online_cpumask(d->cpupool);
 
     spin_lock(&d->node_affinity_lock);
 
     for_each_vcpu ( d, v )
-        cpumask_or(cpumask, cpumask, v->cpu_affinity);
+    {
+        cpumask_and(online_affinity, v->cpu_affinity, online);
+        cpumask_or(cpumask, cpumask, online_affinity);
+    }
 
     for_each_online_node ( node )
         if ( cpumask_intersects(&node_to_cpumask(node), cpumask) )
@@ -354,6 +367,7 @@ void domain_update_node_affinity(struct domain *d)
     d->node_affinity = nodemask;
     spin_unlock(&d->node_affinity_lock);
 
+    free_cpumask_var(online_affinity);
     free_cpumask_var(cpumask);
 }
 

xen/common/schedule.c

@@ -280,12 +280,13 @@ int sched_move_domain(struct domain *d, struct cpupool *c)
 
         SCHED_OP(VCPU2OP(v), insert_vcpu, v);
     }
-    domain_update_node_affinity(d);
 
     d->cpupool = c;
     SCHED_OP(DOM2OP(d), free_domdata, d->sched_priv);
     d->sched_priv = domdata;
 
+    domain_update_node_affinity(d);
+
     domain_unpause(d);
 
     xfree(vcpu_priv);
@@ -535,7 +536,6 @@ int cpu_disable_scheduler(unsigned int cpu)
     struct cpupool *c;
     cpumask_t online_affinity;
     int    ret = 0;
-    bool_t affinity_broken;
 
     c = per_cpu(cpupool, cpu);
     if ( c == NULL )
@@ -543,8 +543,6 @@ int cpu_disable_scheduler(unsigned int cpu)
 
     for_each_domain_in_cpupool ( d, c )
     {
-        affinity_broken = 0;
-
         for_each_vcpu ( d, v )
         {
             vcpu_schedule_lock_irq(v);
@@ -556,7 +554,6 @@ int cpu_disable_scheduler(unsigned int cpu)
                 printk("Breaking vcpu affinity for domain %d vcpu %d\n",
                         v->domain->domain_id, v->vcpu_id);
                 cpumask_setall(v->cpu_affinity);
-                affinity_broken = 1;
             }
 
             if ( v->processor == cpu )
@@ -580,8 +577,7 @@ int cpu_disable_scheduler(unsigned int cpu)
                 ret = -EAGAIN;
         }
 
-        if ( affinity_broken )
-            domain_update_node_affinity(d);
+        domain_update_node_affinity(d);
     }
 
     return ret;