Il 11/04/2013 19:14, Peter Maydell ha scritto: > On 11 April 2013 18:09, Paolo Bonzini <pbonz...@redhat.com> wrote: >> Il 11/04/2013 18:28, Peter Maydell ha scritto: >>> On 11 April 2013 17:10, Paolo Bonzini <pbonz...@redhat.com> wrote: >>>>> Signed-off-by: Paolo Bonzini <pbonz...@redhat.com> >>> This doesn't look right. The MemoryRegion system isn't >>> hw-specific, it's a part of the basic QEMU emulation >>> system which provides functionality to hw/ and other >>> things (like cputlb.c). >> >> The accelerator- and target-independent parts of the basic emulation are >> already in hw/core (not much really, but consider that CPUs are device >> and depend on hw/core/qdev.c). The memory API is simply the interface >> between the accelerators and hw/ (boards & device models). > > Yes, so it should be provided in the place we put our accelerator > implementation (ie .): it is functionality and interface exposed > *to* the code in hw/, not functionality and interface provided > *by* hw/.
Ok, the historical practice was that qdev core was in hw/, and that's what I tried to follow. It makes sense either way to me. But then patch 11 also has to be dropped, otherwise it doesn't make sense. Michael, what do you think? Paolo