[Qemu-devel] [RFC v3 0/3] prebuild cpu QOM tree /machine/node/socket/core ->link-cpu

[chen.fan.fnst]

From: "chen.fan.fnst" <address@hidden>

at present, after hotplug a discontinuous cpu id on source, then done migration,
on target, it will fail to add the unoccupied cpu id which was skipped at 
source,
this cause is on target Qemu prebuild CPU with continuous cpu_index. so after
migration, the cpu infrastructure bewteen source and target are different.

I suppose we could use apic_id as instance_id which was used at registering 
vmstate
when create cpu. on target, we prebuild the specified cpu topology using comand 
line:
 -device /machine/node[]/socket[]/core[]/cpu[], then migration, we could keep 
the same
cpu infrastructure on both side.

V3: get rid of thread object and tie link<cpu> to <core> directly. and prebuild 
full
 core[] and thread[] as init socket[] according to smp_cores and smp_threads.


TODO:
  1. add cpu "path" property which used for specifying the QOM path. 
  2. add -device cpu-foo.path supported.
  3. then we could introduce hot-remove cpu probably.

  I don't know wether this way is right or not. pls tell me. :)

chen.fan.fnst (3):
  cpu: introduce CpuTopoInfo structure for argument simplification
  i386: use CpuTopoInfo instead apic_id as argument for pc_new_cpu()
  i386: introduce cpu QOM hierarchy tree

 hw/i386/pc.c               |  15 +++-
 target-i386/Makefile.objs  |   2 +-
 target-i386/cpu-topology.c | 198 +++++++++++++++++++++++++++++++++++++++++++++
 target-i386/cpu-topology.h |  71 ++++++++++++++++
 target-i386/cpu.c          |  15 +++-
 target-i386/cpu.h          |   6 ++
 target-i386/topology.h     |  33 ++++----
 7 files changed, 317 insertions(+), 23 deletions(-)
 create mode 100644 target-i386/cpu-topology.c
 create mode 100644 target-i386/cpu-topology.h

-- 
1.8.1.4