s390/smp: Switch pcpu_devices to percpu

In preparation of moving the CIF flags from lowcore to pcpu_devices,
convert the pcpu_devices array to use the percpu infrastructure.
This is required because using the pcpu_devices array as it is would
introduce a performance penalty due to the fact that CPU flags for
multiple CPUs would end up in the same cacheline.

Note that a pointer to the pcpu struct of the IPL CPU is still required.
This is because a restart interrupt can be triggered on an offline CPU.
s390 stores the percpu offset in lowcore, but offline CPUs have no
lowcore area allocated. So percpu data cannot be used from an offline
CPU and we need to get the pcpu pointer for the IPL cpu from somewhere
else.
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>