On 8/7/2013 10:42 AM, Jan Beulich wrote:
On 07.08.13 at 17:31, Suravee Suthikulanit <suravee.suthikulpanit@amd.com> wrote:

On 8/7/2013 2:33 AM, Jan Beulich wrote:

On 07.08.13 at 04:40, <suravee.suthikulpanit@amd.com> wrote:

+        /* Filter the bridge devices */
+        if ( (pdev->type == DEV_TYPE_PCIe_BRIDGE)
+        ||   (pdev->type == DEV_TYPE_PCIe2PCI_BRIDGE)
+        ||   (pdev->type == DEV_TYPE_LEGACY_PCI_BRIDGE)
+        ||   (pdev->type == DEV_TYPE_PCI_HOST_BRIDGE) )
+        {
+            AMD_IOMMU_DEBUG("Skipping device %04x:%02x:%02x.%u (type %x)\n",
+                        pdev->seg, PCI_BUS(bdf), PCI_SLOT(bdf), PCI_FUNC(bdf),
+                        pdev->type);

I can see why host bridges can be skipped, but is this really also true
for other bridge types, most importantly legacy ones?

I am using the same logic here as in Intel's version in the 
driver/passthrough/vtd/iommu/domain_context_map.

No, not really: That code establishes a mapping for the upstream
bridge of a legacy PCI device when handling that device. Your
proposed code doesn't afaics. (This I understand is because
bridges aren't expected to get assigned to guests, and hence
otherwise the mapping for the bridge would never get established
for a device behind it for other than Dom0.)

Jan
Jan,

AFAICT from the domain_context_mapping() below, which get called when the devices are assigned to domains

static int domain_context_mapping(
    struct domain *domain, u8 devfn, const struct pci_dev *pdev)
{
    struct acpi_drhd_unit *drhd;
    int ret = 0;
    u8 seg = pdev->seg, bus = pdev->bus, secbus;

    drhd = acpi_find_matched_drhd_unit(pdev);
    if ( !drhd )
        return -ENODEV;

    ASSERT(spin_is_locked(&pcidevs_lock));

    switch ( pdev->type )
    {
    case DEV_TYPE_PCIe_BRIDGE:
    case DEV_TYPE_PCIe2PCI_BRIDGE:
    case DEV_TYPE_LEGACY_PCI_BRIDGE:
        break;

These bridges are actually not mapped (i.e. skipped).  That is why I think the logic above is supposed to be doing the same thing.

Suravee