xen/sched: introduce unit_runnable_state()

Today the vcpu runstate of a new scheduled vcpu is always set to
"running" even if at that time vcpu_runnable() is already returning
false due to a race (e.g. with pausing the vcpu).

With core scheduling this can no longer work as not all vcpus of a
schedule unit have to be "running" when being scheduled. So the vcpu's
new runstate has to be selected at the same time as the runnability of
the related schedule unit is probed.

For this purpose introduce a new helper unit_runnable_state() which
will save the new runstate of all tested vcpus in a new field of the
vcpu struct.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Dario Faggioli <dfaggioli@suse.com>

xen/common/domain.c

@@ -157,6 +157,7 @@ struct vcpu *vcpu_create(struct domain *d, unsigned int vcpu_id)
     if ( is_idle_domain(d) )
     {
         v->runstate.state = RUNSTATE_running;
+        v->new_state = RUNSTATE_running;
     }
     else
     {

xen/common/sched_arinc653.c

@@ -563,7 +563,7 @@ a653sched_do_schedule(
     if ( !((new_task != NULL)
            && (AUNIT(new_task) != NULL)
            && AUNIT(new_task)->awake
-           && unit_runnable(new_task)) )
+           && unit_runnable_state(new_task)) )
         new_task = IDLETASK(cpu);
     BUG_ON(new_task == NULL);
 

xen/common/sched_credit.c

@@ -1894,7 +1894,7 @@ static void csched_schedule(
     if ( !test_bit(CSCHED_FLAG_UNIT_YIELD, &scurr->flags)
          && !tasklet_work_scheduled
          && prv->ratelimit
-         && unit_runnable(unit)
+         && unit_runnable_state(unit)
          && !is_idle_unit(unit)
          && runtime < prv->ratelimit )
     {
@@ -1939,33 +1939,36 @@ static void csched_schedule(
         dec_nr_runnable(sched_cpu);
     }
 
-    snext = __runq_elem(runq->next);
-
-    /* Tasklet work (which runs in idle UNIT context) overrides all else. */
-    if ( tasklet_work_scheduled )
-    {
-        TRACE_0D(TRC_CSCHED_SCHED_TASKLET);
-        snext = CSCHED_UNIT(sched_idle_unit(sched_cpu));
-        snext->pri = CSCHED_PRI_TS_BOOST;
-    }
-
     /*
      * Clear YIELD flag before scheduling out
      */
     clear_bit(CSCHED_FLAG_UNIT_YIELD, &scurr->flags);
 
-    /*
-     * SMP Load balance:
-     *
-     * If the next highest priority local runnable UNIT has already eaten
-     * through its credits, look on other PCPUs to see if we have more
-     * urgent work... If not, csched_load_balance() will return snext, but
-     * already removed from the runq.
-     */
-    if ( snext->pri > CSCHED_PRI_TS_OVER )
-        __runq_remove(snext);
-    else
-        snext = csched_load_balance(prv, sched_cpu, snext, &migrated);
+    do {
+        snext = __runq_elem(runq->next);
+
+        /* Tasklet work (which runs in idle UNIT context) overrides all else. */
+        if ( tasklet_work_scheduled )
+        {
+            TRACE_0D(TRC_CSCHED_SCHED_TASKLET);
+            snext = CSCHED_UNIT(sched_idle_unit(sched_cpu));
+            snext->pri = CSCHED_PRI_TS_BOOST;
+        }
+
+        /*
+         * SMP Load balance:
+         *
+         * If the next highest priority local runnable UNIT has already eaten
+         * through its credits, look on other PCPUs to see if we have more
+         * urgent work... If not, csched_load_balance() will return snext, but
+         * already removed from the runq.
+         */
+        if ( snext->pri > CSCHED_PRI_TS_OVER )
+            __runq_remove(snext);
+        else
+            snext = csched_load_balance(prv, sched_cpu, snext, &migrated);
+
+    } while ( !unit_runnable_state(snext->unit) );
 
     /*
      * Update idlers mask if necessary. When we're idling, other CPUs

xen/common/sched_credit2.c

@@ -3291,7 +3291,7 @@ runq_candidate(struct csched2_runqueue_data *rqd,
      * In fact, it may be the case that scurr is about to spin, and there's
      * no point forcing it to do so until rate limiting expires.
      */
-    if ( !yield && prv->ratelimit_us && unit_runnable(scurr->unit) &&
+    if ( !yield && prv->ratelimit_us && unit_runnable_state(scurr->unit) &&
          (now - scurr->unit->state_entry_time) < MICROSECS(prv->ratelimit_us) )
     {
         if ( unlikely(tb_init_done) )
@@ -3345,7 +3345,7 @@ runq_candidate(struct csched2_runqueue_data *rqd,
      *
      * Of course, we also default to idle also if scurr is not runnable.
      */
-    if ( unit_runnable(scurr->unit) && !soft_aff_preempt )
+    if ( unit_runnable_state(scurr->unit) && !soft_aff_preempt )
         snext = scurr;
     else
         snext = csched2_unit(sched_idle_unit(cpu));
@@ -3405,7 +3405,8 @@ runq_candidate(struct csched2_runqueue_data *rqd,
          * some budget, then choose it.
          */
         if ( (yield || svc->credit > snext->credit) &&
-             (!has_cap(svc) || unit_grab_budget(svc)) )
+             (!has_cap(svc) || unit_grab_budget(svc)) &&
+             unit_runnable_state(svc->unit) )
             snext = svc;
 
         /* In any case, if we got this far, break. */

xen/common/sched_null.c

@@ -864,7 +864,8 @@ static void null_schedule(const struct scheduler *ops, struct sched_unit *prev,
             cpumask_set_cpu(sched_cpu, &prv->cpus_free);
     }
 
-    if ( unlikely(prev->next_task == NULL || !unit_runnable(prev->next_task)) )
+    if ( unlikely(prev->next_task == NULL ||
+                  !unit_runnable_state(prev->next_task)) )
         prev->next_task = sched_idle_unit(sched_cpu);
 
     NULL_UNIT_CHECK(prev->next_task);

xen/common/sched_rt.c

@@ -1092,12 +1092,18 @@ rt_schedule(const struct scheduler *ops, struct sched_unit *currunit,
     else
     {
         snext = runq_pick(ops, cpumask_of(sched_cpu));
+
         if ( snext == NULL )
             snext = rt_unit(sched_idle_unit(sched_cpu));
+        else if ( !unit_runnable_state(snext->unit) )
+        {
+            q_remove(snext);
+            snext = rt_unit(sched_idle_unit(sched_cpu));
+        }
 
         /* if scurr has higher priority and budget, still pick scurr */
         if ( !is_idle_unit(currunit) &&
-             unit_runnable(currunit) &&
+             unit_runnable_state(currunit) &&
              scurr->cur_budget > 0 &&
              ( is_idle_unit(snext->unit) ||
                compare_unit_priority(scurr, snext) > 0 ) )

xen/common/schedule.c

@@ -280,7 +280,7 @@ static inline void sched_unit_runstate_change(struct sched_unit *unit,
     for_each_sched_unit_vcpu ( unit, v )
     {
         if ( running )
-            vcpu_runstate_change(v, RUNSTATE_running, new_entry_time);
+            vcpu_runstate_change(v, v->new_state, new_entry_time);
         else
             vcpu_runstate_change(v,
                 ((v->pause_flags & VPF_blocked) ? RUNSTATE_blocked :

xen/include/xen/sched-if.h

@@ -93,6 +93,36 @@ static inline bool unit_runnable(const struct sched_unit *unit)
     return false;
 }
 
+/*
+ * Returns whether a sched_unit is runnable and sets new_state for each of its
+ * vcpus. It is mandatory to determine the new runstate for all vcpus of a unit
+ * without dropping the schedule lock (which happens when synchronizing the
+ * context switch of the vcpus of a unit) in order to avoid races with e.g.
+ * vcpu_sleep().
+ */
+static inline bool unit_runnable_state(const struct sched_unit *unit)
+{
+    struct vcpu *v;
+    bool runnable, ret = false;
+
+    if ( is_idle_unit(unit) )
+        return true;
+
+    for_each_sched_unit_vcpu ( unit, v )
+    {
+        runnable = vcpu_runnable(v);
+
+        v->new_state = runnable ? RUNSTATE_running
+                                : (v->pause_flags & VPF_blocked)
+                                  ? RUNSTATE_blocked : RUNSTATE_offline;
+
+        if ( runnable )
+            ret = true;
+    }
+
+    return ret;
+}
+
 static inline void sched_set_res(struct sched_unit *unit,
                                  struct sched_resource *res)
 {

xen/include/xen/sched.h

@@ -174,6 +174,7 @@ struct vcpu
         XEN_GUEST_HANDLE(vcpu_runstate_info_compat_t) compat;
     } runstate_guest; /* guest address */
 #endif
+    unsigned int     new_state;
 
     /* Has the FPU been initialised? */
     bool             fpu_initialised;