On 2017年04月20日 15:04, Peter Xu wrote: > On Thu, Apr 20, 2017 at 06:36:16AM +0000, Liu, Yi L wrote: > > [...] > >>>> In my understanding, container->space->as->root cannot work here no >>>> matter passthru-mode is enabled or not. The code here is aiming to >>>> check if vIOMMU exists. After the vfio series, the vtd_dev_as->root is >>>> not initialized to be a iommu MemoryRegion. Compared with checking if >>>> it is system_memory(), I think adding a mechanism to get the iommu >>> MemoryRegion may be a better choice. Just like the current >>> pci_device_iommu_address_space(). >>> >>> Isn't pci_device_iommu_address_space() used to get that IOMMU memory region? > > Again I should say s/memory region/address space/... > >> >> It actually returns the AddressSpace, and the AddressSpace includes a memory >> region. >> It is as->root. But after adding the vfio series, through the IOMMU memory >> region >> is got, but it has no iommu_ops. Just as the following code shows. That's >> why I said >> even without passthru-mode, Tianyu's that code snippet is not able to get >> the correct >> check. >> >> memory_region_init(&vtd_dev_as->root, OBJECT(s), >> "vtd_root", UINT64_MAX); >> address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, name); > > The problem is, I am not sure whether there is always _only_ one IOMMU > region behind one device. E.g., IIUC ppc can have more than one IOMMU > memory regions, but translate for different address ranges. >
If we really want to check this via memory region, we may check all subregions under container's address space. Register vIOMMU related notifiers if one of them has iommu_ops, -- Best regards Tianyu Lan
