On 02/22/2012 03:26 PM, Peter Maydell wrote:
On 22 February 2012 10:15, Igor Mitsyanko<i.mitsya...@samsung.com> wrote:
This patchset cleans up and optimizes vmstate implementation.
Patch 1 is a trivial bug fixing.
Patches 2 and 3 replaces target_phys_addr_t in pxa implementation
to uint32_t.
Patch 4 moves VMSTATE_UINTTL from hw.h to vmstate.h. Explicit dependency
on NEED_CPU_H is droped, I failed to understand why it was presented at all.
So if we apply patches 1-3 (which all look plausible) then the only
remaining user of VMSTATE_UINTTL is target-i386/machine.c as far as
I can see.
This leaves me wondering if we shouldn't just put it actually in
target-i386/machine.c as a convenience macro for that specific CPU
to avoid having to have more #ifdef TARGET_X86_64s. (I note that
the machine.c code is already pretty inconsistent, eg lstar and
cstar are defined as target_ulong and saved with VMSTATE_UINT64.)
Basically VMSTATE_UINTTL seems like a bit of a dangerous thing to
leave lying around as there aren't really very many use cases
for it...
I personally don't really like all these "hack" VMSTATE macros spread
all around QEMU code. I would prefer to have all convenient VMSTATE_*s
in one place and eventually replace all file-specific hack macro with
standard ones. Not sure that it's worth an effort though.
If we're going to move UINTTL* to target-i386/machine.c, then they
probably should be implemented in the same way as they are implemented
in hw.h now. But without NEED_CPU_H.
--
Mitsyanko Igor
ASWG, Moscow R&D center, Samsung Electronics
email: i.mitsya...@samsung.com
