On Thu, Jun 02, 2016 at 11:15:54PM +0300, Marcel Apfelbaum wrote:
[...]
static void vtd_realize(DeviceState *dev, Error **errp)
{
+ PCIBus *bus = PC_MACHINE(qdev_get_machine())->bus;
IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
VTD_DPRINTF(GENERAL, "");
@@ -2029,6 +2043,9 @@ static void vtd_realize(DeviceState *dev, Error **errp)
s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash,
vtd_uint64_equal,
g_free, g_free);
vtd_init(s);
+ sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
+ bus->iommu_fn = vtd_host_dma_iommu;
+ bus->iommu_opaque = dev;
Here, shall we still use pci_setup_iommu() to keep the two fields
private for pci framework?
Btw, I am rebasing Intel IR work onto this patchset, but encountered
issues (guest hang, or errornous interrupts) when guest specify more
than 1 vcpus (everything is cool as long as vcpu=1). Maybe there is
something wrong during the rebase, still investigating. Please shoot
if there is any clue.