On Wed, Feb 08, 2023 at 08:31:45PM -0800, Yi Liu wrote:
Today iommufd allocates the kernel-managed hw_pagetabe implicitly when device is attached to an IOAS. This links the hw_pagetable to IOPT within IOAS.
However, this is not the perfect way. It makes much sense to let userspace explicitly request hw_pagtable allocation via iommufd. The reason is even though the hw_pagetable is kernel-managed, the mappings are feed by userspace. Also, this makes the lifecircle of kernel-managed hw_pagetable more clear during usage. This is very important in the usage of nested translation, in which the kernel-managed hw_pagetable would be used as the stage-2 hw_pagetable. In such case, both stage-1 and stage-2 hw_pagetable should be allocated by userspace to ensure the life-circle.
This adds an ioctl IOMMU_HWPT_ALLOC for the hw_pagetable allocation. For kernel-managed hw_pagetable, userspace should provide an IOAS ID in the allocation request.
Signed-off-by: Nicolin Chen nicolinc@nvidia.com Signed-off-by: Yi Liu yi.l.liu@intel.com
drivers/iommu/iommufd/device.c | 11 ++- drivers/iommu/iommufd/hw_pagetable.c | 116 ++++++++++++++++++++++++ drivers/iommu/iommufd/iommufd_private.h | 15 +++ drivers/iommu/iommufd/main.c | 3 + include/uapi/linux/iommufd.h | 48 ++++++++++ 5 files changed, 191 insertions(+), 2 deletions(-)
This patch and its requirements should all be first in the series. A mini series who's only job is to add IOMMU_HWPT_ALLOC
Then patches to add IOMMU_HWPT_INVALIDATE
Then the vt-d implementation of all this, including the vt-d specific changes to the uapi/etc.
Jason