Re: [Qemu-devel] [PATCH 1/2] i386/pc: propagate flash size from pc_system_flash_init() to pc_init1()

[Peter Maydell]

On 8 November 2013 15:27, Laszlo Ersek <address@hidden> wrote:
> On 11/08/13 16:16, Peter Maydell wrote:
>> That said, having to pass the size of a sub-sub-region
>> all the way back up the call stack is very odd and makes
>> me wonder if it's really the right way to do it...
>> The top level shouldn't have to care like that about
>> details of the bottom of the callstack.
>
> I agree. It's just that system.flash and pci-hole are siblings in the
> same container (they shouldn't overlap, *or* they should have clearly
> different priorities between them). We have two call chains rooted in
> pc_init1(), and the "ends" of those chains need to coordinate with each
> other (they set up the two regions, respectively, and both need the
> boundary between them). We could introduce a new global, but that's not
> exactly a step forward :)

I suspect the correct approach for the pci-hole is to
use priorities. (Also possibly to refactor the code so
that the things creating memory regions directly in the
top level container aren't four levels down the callstack.
The natural thing would be for the callgraph to be vaguely
parallel to memory region hierarchy, so all the code
managing a particular container region is in the same
place in the source rather than distributed.)

-- PMM