[PATCH] x86: make "dom0_nodes=" work with credit2

[Jan Beulich <jbeulich@xxxxxxxx>]

Apart from setting the affinities, vCPU-s / units also need to be
migrated off of CPUs which aren't part of the affinity. Otherwise, for a
reason I haven't been able to determine, credit2 won't actually schedule
them, resulting in the Dom0 boot process to hang (either right away or
when bringing up "secondary" vCPU-s).

Fixes: dafd936ddd ("Make credit2 the default scheduler")
Reported-by: Olaf Hering <ohering@xxxxxxx>
Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
The Fixes: tag isn't very precise - it's rather the commit exposing the
issue by default. I haven't been able to identify the actual commit
which did introduce the problem; it may well be that it has always been
there since the introduction of credit2.

Credit2 moving the vCPU-s off of their initially assigned ones right
away of course renders sched_select_initial_cpu()'s use of cpu_cycle()
pretty useless. But I guess that's still useful for other schedulers.
I wonder though whether sched_init_vcpu() shouldn't use the CPU map
calculated by sched_select_initial_cpu() for its call to
sched_set_affinity() in the non-pinned case, rather than setting "all".
(I guess doing so might mask the issue at hand, but I think the change
here would still be applicable at least from an abstract pov.)

--- a/xen/common/sched/core.c
+++ b/xen/common/sched/core.c
@@ -3403,9 +3403,15 @@ void __init sched_setup_dom0_vcpus(struc
     {
         for_each_sched_unit ( d, unit )
         {
+            spinlock_t *lock = unit_schedule_lock_irq(unit);
+
             if ( !opt_dom0_vcpus_pin && !dom0_affinity_relaxed )
                 sched_set_affinity(unit, &dom0_cpus, NULL);
             sched_set_affinity(unit, NULL, &dom0_cpus);
+
+            sched_unit_migrate_start(unit);
+            unit_schedule_unlock_irq(lock, unit);
+            sched_unit_migrate_finish(unit);
         }
     }