On Fri, 2015-07-10 at 20:43 +1000, Alexey Kardashevskiy wrote: > In some cases PCI BARs are registered as RAM via > memory_region_init_ram_ptr() and the vfio_memory_listener will be called > on them too. However DMA will not be performed to/from these regions so > just skip them.
Who says? What about peer-to-peer DMA? We have all sorts of FUD about whether the hardware handles this correctly, but mapping PCI MMIO BARs into the IOMMU allows the possibility of supporting p2p DMA between devices in the guest. Thanks, Alex > > Signed-off-by: Alexey Kardashevskiy <a...@ozlabs.ru> > --- > hw/vfio/common.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/hw/vfio/common.c b/hw/vfio/common.c > index d115ec9..225cdc7 100644 > --- a/hw/vfio/common.c > +++ b/hw/vfio/common.c > @@ -248,7 +248,8 @@ static bool > vfio_listener_skipped_section(MemoryRegionSection *section) > * are never accessed by the CPU and beyond the address width of > * some IOMMU hardware. TODO: VFIO should tell us the IOMMU > width. > */ > - section->offset_within_address_space & (1ULL << 63); > + section->offset_within_address_space & (1ULL << 63) || > + memory_region_is_skip_dump(section->mr); > } > > static void vfio_iommu_map_notify(Notifier *n, void *data)
