Am 04.05.2012 18:16, schrieb Andreas Färber: > Am 04.05.2012 12:54, schrieb Juan Quintela: >> This makes several changes: >> - exports VMStateDescription vmstate_cpu non-static. >> - makes sure that every cpu has a vmstate_cpu or cpu_save/load defined >> - for the architecture that had nothing, it just register the cpu as >> unmigratable. >> - Depending on CPU_SAVE_VERSION we register old/new migration style >> - Add copyrights to the new files >> >> Signed-off-by: Juan Quintela <quint...@redhat.com> >> --- >> exec.c | 2 ++ >> target-alpha/machine.c | 12 +----------- >> target-i386/machine.c | 2 +- >> target-lm32/machine.c | 2 +- >> target-m68k/machine.c | 21 +++++++++++++++++++++ >> target-s390x/machine.c | 14 ++++++-------- >> target-sh4/machine.c | 21 +++++++++++++++++++++ >> target-xtensa/machine.c | 14 ++++++-------- >> vmstate.h | 2 ++ >> 9 files changed, 61 insertions(+), 29 deletions(-) [...] > We have QOM CPUState now, so please add a vmsd field to CPUClass instead > of exporting global vmstate_cpu variables.
Thinking more about this, I see that we will need to assign it in cpu.c:xxx_class_init() (translate_init.c for ppc) as opposed to machine.c. The declaration still shouldn't go into vmstate.h so that it remains a target-local decision how to model this (think CPU subclasses). I'd hate to #include "machine.c" in cpu.c. Having the declaration in cpu.h is the least intrusive solution I see. Moving the code into cpu.c would match qdev but would bloat the files. What do you think? > It then matches what we do > for former qdev devices and will in the future allow to compile this > code only twice. BTW the easiest way to address my comment on 01/35 would be to simply squash the two commits. :) Andreas -- SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg
