A recent change in the hypervisor memory allocation framework led to
crashes when unplugging host cpus.

This was due to the (correct) assertion that allocating and freeing
memory is allowed with enabled interrupts only. As the main cpu unplug
operation is done in stop-machine context, this assertion triggers in
debug builds.

Correct that by pre-allocating all needed memory while interrupts are
still on, and free memory after interrupts are enabled again.

Juergen Gross (3):
  xen/sched: introduce cpupool_update_node_affinity()
  xen/sched: carve out memory allocation and freeing from
    schedule_cpu_rm()
  xen/sched: fix cpu hotplug

 xen/common/sched/core.c    | 198 +++++++++++++++++++++----------------
 xen/common/sched/cpupool.c | 119 +++++++++++++++++-----
 xen/common/sched/private.h |  21 +++-
 3 files changed, 229 insertions(+), 109 deletions(-)

-- 
2.35.3