Il 04/06/2013 16:56, Peter Maydell ha scritto: >>> As a concrete example, if somebody submitted cirrus_vga >>> as a new driver, I have no idea how to tell that it needs >>> to set the owner for its memory regions, when 99% of >>> other devices don't. I think this is going to result in >>> "forgot to set owner" bugs. >> >> Because cirrus is adding regions directly to address_space_memory/io. >> As documented: >> >> * The device must set the owner itself >> * only if it uses memory_region_add_subregion directly on some address >> * space, or after the parent region is passed to the bus (for example >> * dynamically while the device runs). > > OK, so why doesn't your patchset make the places in > hw/arm/omap1.c which add memory regions directly > to a subregion set the owner of the region?
Because these aren't qdevified. > (or any of the many other places where we do similar things). Note that it's only necessary to do so when you add those to the address space, not to other regions. Paolo
