On Mon, Apr 26, 2010 at 12:54 PM, Anthony Liguori <anth...@codemonkey.ws> wrote: > On 04/26/2010 01:49 PM, Marcelo Tosatti wrote: >> >> On Mon, Apr 26, 2010 at 01:27:30PM -0500, Anthony Liguori wrote: >> >>> >>> On 04/26/2010 12:59 PM, Marcelo Tosatti wrote: >>> >>>> >>>> Which allows drivers to register an mmaped region into ram block >>>> mappings. >>>> To be used by device assignment driver. >>>> >>> >>> This doesn't make much sense to me. >>> >>> Do you use this like: >>> >>> qemu_ram_map(64k, ptr); >>> assert(qemu_ram_alloc(64k) == ptr); >>> >> >> No. hw/device-assignment.c in qemu-kvm mmaps >> /sys/bus/pci/devices/x:y:z/resourcen (the PCI devices memory regions) to >> the guest. >> > > I understand, but how do you use qemu_ram_map() to actually map that memory > to a given PCI device resource? I assume you rely on it getting put on the > front of the list so that the next qemu_ram_alloc() will be at that > location.
In my shared memory patch, I passed the offset returned from qemu_ram_mmap to cpu_register_physical_memory from within the map function passed to pci_register_bar. Could the same not be done? Is there something incorrect with this approach? > > Regards, > > Anthony Liguori > >>> If so, I think this is not the best API. I'd rather see >>> qemu_ram_map() register a symbolic name for the region and for there >>> to be a qemu_ram_alloc() variant that allocated by name. >>> >>> Regards, >>> >>> Anthony Liguori >>> >>> > >
