Re: [Qemu-devel] [PATCH 4/5] integratorcp: convert integratorcm to VMState

[Peter Maydell]

On 9 November 2011 15:05, Avi Kivity <address@hidden> wrote:
> On 11/09/2011 04:40 PM, Anthony Liguori wrote:
>>
>> typedef struct {
>>     SysBusDevice busdev;
>>     uint32_t memsz;
>>     MemoryRegion flash;
>>     bool flash_mapped;
>
> Both flash.has_parent and flash_mapped are slaved to a bit of one of the
> registers below.

Yes. The only reason for having a separate flash_mapped would be if
it was *not* saved/loaded and we needed it to track whether the
memory region API thought the region was mapped or not (ie whether
calling memory_region_add_subregion_overlap() from the post-load-hook
would assert, which is what it does in master currently.)

This is why I'm pushing back on Benoit's patch here -- I don't
want us to carry around this extra flash_mapped flag unless we
really need it.

>From a really quick glance at Avi's memory/mutators branch I think
the long term plan would be that rather than mapping/unmapping the
region we just call memory_region_set_enabled() which would cope
happily with being asked to enable an already-enabled region.
Is that right?

-- PMM