This small series moves the X86CPU object creation to the cpu.c code, as the object creation will depend on the CPU class lookup, that will be done inside cpu.c.
I expect this to make the x86 CPU subclass work simpler. Maybe it will help simplify some of the CPU properties work, too. This version is simply a rebase of v1 on top of Igor's series: Subject: [Qemu-devel] [PATCH 0/6] x86 CPU cleanup (wave 2) Git branch for testing: git://github.com/ehabkost/qemu-hacks.git cpu-x86-create.v2 https://github.com/ehabkost/qemu-hacks/tree/cpu-x86-create.v2 As Igor have suggested instead moving cpu_x86_init() to cpu.c, I will try to explain the rationale, here: The CPU creation consists of multiple steps: 1) Splitting cpu_model string into CPU model name and feature string; 2) Choosing the CPU model class; 3) Creating CPU object; 4) Parsing feature string; 5) cpu_x86_realize() And: * cpu_x86_register() already does steps 1 and 4. * cpu_x86_init() does steps 2, 3, and 5 * We need to separate step 5 from the rest, as step 5 will eventually become the realize() method of the CPU object. To accomplish that, the first patch in this series simply moves 3 lines of code (steps 2 and 3) from cpu_x86_init() to cpu_x86_register(), and renames cpu_x86_register() to cpu_x86_create(), so now: * cpu_x86_create() does steps 1 to 4 * cpu_x86_init() simply become a trivial wrapper to call cpu_x86_create() (steps 1-4) and cpu_x86_realize() (step 5) Eduardo Habkost (2): target-i386: move CPU object creation to cpu.c target-i386: make cpu_x86_create() get Error argument target-i386/cpu.c | 22 ++++++++++++++++++---- target-i386/cpu.h | 2 +- target-i386/helper.c | 26 ++++++++++++++------------ 3 files changed, 33 insertions(+), 17 deletions(-) -- 1.7.11.7
