Reflect cpupool in numa node affinity (v5)

diff -r 370924e204dc -r 5589314aa984 xen/common/domctl.c
@@ -518,7 +518,7 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
             goto maxvcpu_out;
 
         ret = -ENOMEM;
-        online = (d->cpupool == NULL) ? &cpu_online_map : d->cpupool->cpu_valid;
+        online = cpupool_online_cpumask(d->cpupool);
         if ( max > d->max_vcpus )
         {
             struct vcpu **vcpus;

diff -r 370924e204dc -r 5589314aa984 xen/common/sched_credit.c
@@ -72,8 +72,6 @@
 #define CSCHED_VCPU(_vcpu)  ((struct csched_vcpu *) (_vcpu)->sched_priv)
 #define CSCHED_DOM(_dom)    ((struct csched_dom *) (_dom)->sched_priv)
 #define RUNQ(_cpu)          (&(CSCHED_PCPU(_cpu)->runq))
-#define CSCHED_CPUONLINE(_pool)    \
-    (((_pool) == NULL) ? &cpupool_free_cpus : (_pool)->cpu_valid)
 
 
 /*

@@ -471,7 +469,7 @@ _csched_cpu_pick(const struct scheduler 
      * Pick from online CPUs in VCPU's affinity mask, giving a
      * preference to its current processor if it's in there.
      */
-    online = CSCHED_CPUONLINE(vc->domain->cpupool);
+    online = cpupool_scheduler_cpumask(vc->domain->cpupool);
     cpumask_and(&cpus, online, vc->cpu_affinity);
     cpu = cpumask_test_cpu(vc->processor, &cpus)
             ? vc->processor

@@ -1230,7 +1228,7 @@ csched_load_balance(struct csched_privat
     int peer_cpu;
 
     BUG_ON( cpu != snext->vcpu->processor );
-    online = CSCHED_CPUONLINE(per_cpu(cpupool, cpu));
+    online = cpupool_scheduler_cpumask(per_cpu(cpupool, cpu));
 
     /* If this CPU is going offline we shouldn't steal work. */
     if ( unlikely(!cpumask_test_cpu(cpu, online)) )

diff -r 370924e204dc -r 5589314aa984 xen/common/sched_credit2.c
@@ -169,8 +169,6 @@ integer_param("sched_credit2_migrate_res
     ((struct csched_private *)((_ops)->sched_data))
 #define CSCHED_VCPU(_vcpu)  ((struct csched_vcpu *) (_vcpu)->sched_priv)
 #define CSCHED_DOM(_dom)    ((struct csched_dom *) (_dom)->sched_priv)
-#define CSCHED_CPUONLINE(_pool)    \
-    (((_pool) == NULL) ? &cpupool_free_cpus : (_pool)->cpu_valid)
 /* CPU to runq_id macro */
 #define c2r(_ops, _cpu)     (CSCHED_PRIV(_ops)->runq_map[(_cpu)])
 /* CPU to runqueue struct macro */

diff -r 370924e204dc -r 5589314aa984 xen/common/sched_sedf.c
@@ -12,9 +12,6 @@
 #include <xen/softirq.h>
 #include <xen/time.h>
 #include <xen/errno.h>
-
-#define SEDF_CPUONLINE(_pool)                                             \
-    (((_pool) == NULL) ? &cpupool_free_cpus : (_pool)->cpu_valid)
 
 #ifndef NDEBUG
 #define SEDF_STATS

@@ -397,7 +394,7 @@ static int sedf_pick_cpu(const struct sc
     cpumask_t online_affinity;
     cpumask_t *online;
 
-    online = SEDF_CPUONLINE(v->domain->cpupool);
+    online = cpupool_scheduler_cpumask(v->domain->cpupool);
     cpumask_and(&online_affinity, v->cpu_affinity, online);
     return cpumask_first(&online_affinity);
 }

@@ -801,7 +798,8 @@ static struct task_slice sedf_do_schedul
      */
     if ( tasklet_work_scheduled ||
          (list_empty(runq) && list_empty(waitq)) ||
-         unlikely(!cpumask_test_cpu(cpu, SEDF_CPUONLINE(per_cpu(cpupool, cpu)))) )
+         unlikely(!cpumask_test_cpu(cpu,
+                   cpupool_scheduler_cpumask(per_cpu(cpupool, cpu)))) )
     {
         ret.task = IDLETASK(cpu);
         ret.time = SECONDS(1);

diff -r 370924e204dc -r 5589314aa984 xen/common/schedule.c
@@ -77,9 +77,7 @@ static struct scheduler __read_mostly op
 
 #define DOM2OP(_d)    (((_d)->cpupool == NULL) ? &ops : ((_d)->cpupool->sched))
 #define VCPU2OP(_v)   (DOM2OP((_v)->domain))
-#define VCPU2ONLINE(_v)                                                    \
-         (((_v)->domain->cpupool == NULL) ? &cpu_online_map                \
-         : (_v)->domain->cpupool->cpu_valid)
+#define VCPU2ONLINE(_v) cpupool_online_cpumask((_v)->domain->cpupool)
 
 static inline void trace_runstate_change(struct vcpu *v, int new_state)
 {

@@ -1418,7 +1416,7 @@ void schedule_dump(struct cpupool *c)
     cpumask_t        *cpus;
 
     sched = (c == NULL) ? &ops : c->sched;
-    cpus = (c == NULL) ? &cpupool_free_cpus : c->cpu_valid;
+    cpus = cpupool_scheduler_cpumask(c);
     printk("Scheduler: %s (%s)\n", sched->name, sched->opt_name);
     SCHED_OP(sched, dump_settings);
 

diff -r 370924e204dc -r 5589314aa984 xen/include/xen/sched-if.h
@@ -204,4 +204,9 @@ struct cpupool
     atomic_t         refcnt;
 };
 
+#define cpupool_scheduler_cpumask(_pool) \
+    (((_pool) == NULL) ? &cpupool_free_cpus : (_pool)->cpu_valid)
+#define cpupool_online_cpumask(_pool) \
+    (((_pool) == NULL) ? &cpu_online_map : (_pool)->cpu_valid)
+
 #endif /* __XEN_SCHED_IF_H__ */

diff -r 5589314aa984 -r 0caec9849a8c xen/common/domain.c
@@ -220,6 +220,7 @@ struct domain *domain_create(
     INIT_PAGE_LIST_HEAD(&d->xenpage_list);
 
     spin_lock_init(&d->node_affinity_lock);
+    d->node_affinity = NODE_MASK_ALL;
 
     spin_lock_init(&d->shutdown_lock);
     d->shutdown_code = -1;

@@ -333,23 +334,27 @@ struct domain *domain_create(
 
 void domain_update_node_affinity(struct domain *d)
 {
-    cpumask_t cpumask;
+    cpumask_var_t cpumask;
     nodemask_t nodemask = NODE_MASK_NONE;
     struct vcpu *v;
     unsigned int node;
 
-    cpumask_clear(&cpumask);
+    if ( !zalloc_cpumask_var(&cpumask) )
+        return;
+
     spin_lock(&d->node_affinity_lock);
 
     for_each_vcpu ( d, v )
-        cpumask_or(&cpumask, &cpumask, v->cpu_affinity);
+        cpumask_or(cpumask, cpumask, v->cpu_affinity);
 
     for_each_online_node ( node )
-        if ( cpumask_intersects(&node_to_cpumask(node), &cpumask) )
+        if ( cpumask_intersects(&node_to_cpumask(node), cpumask) )
             node_set(node, nodemask);
 
     d->node_affinity = nodemask;
     spin_unlock(&d->node_affinity_lock);
+
+    free_cpumask_var(cpumask);
 }
 
 

diff -r 0caec9849a8c -r 844f36fd8c76 xen/common/cpupool.c
@@ -220,6 +220,7 @@ static int cpupool_assign_cpu_locked(str
 {
     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(str
         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;
 }
 

diff -r 0caec9849a8c -r 844f36fd8c76 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 @@ void domain_update_node_affinity(struct 
 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 
     d->node_affinity = nodemask;
     spin_unlock(&d->node_affinity_lock);
 
+    free_cpumask_var(online_affinity);
     free_cpumask_var(cpumask);
 }
 

diff -r 0caec9849a8c -r 844f36fd8c76 xen/common/schedule.c
@@ -280,11 +280,12 @@ int sched_move_domain(struct domain *d, 
 
         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);
 

@@ -535,7 +536,6 @@ int cpu_disable_scheduler(unsigned int c
     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 c
 
     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 c
                 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 c
                 ret = -EAGAIN;
         }
 
-        if ( affinity_broken )
-            domain_update_node_affinity(d);
+        domain_update_node_affinity(d);
     }
 
     return ret;